11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/nfs/write.c 31da177e4SLinus Torvalds * 47c85d900STrond Myklebust * Write file data over NFS. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de> 71da177e4SLinus Torvalds */ 81da177e4SLinus Torvalds 91da177e4SLinus Torvalds #include <linux/types.h> 101da177e4SLinus Torvalds #include <linux/slab.h> 111da177e4SLinus Torvalds #include <linux/mm.h> 121da177e4SLinus Torvalds #include <linux/pagemap.h> 131da177e4SLinus Torvalds #include <linux/file.h> 141da177e4SLinus Torvalds #include <linux/writeback.h> 1589a09141SPeter Zijlstra #include <linux/swap.h> 16074cc1deSTrond Myklebust #include <linux/migrate.h> 171da177e4SLinus Torvalds 181da177e4SLinus Torvalds #include <linux/sunrpc/clnt.h> 191da177e4SLinus Torvalds #include <linux/nfs_fs.h> 201da177e4SLinus Torvalds #include <linux/nfs_mount.h> 211da177e4SLinus Torvalds #include <linux/nfs_page.h> 223fcfab16SAndrew Morton #include <linux/backing-dev.h> 23afeacc8cSPaul Gortmaker #include <linux/export.h> 24af7cf057STrond Myklebust #include <linux/freezer.h> 25af7cf057STrond Myklebust #include <linux/wait.h> 261eb5d98fSJeff Layton #include <linux/iversion.h> 273fcfab16SAndrew Morton 287c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 291da177e4SLinus Torvalds 301da177e4SLinus Torvalds #include "delegation.h" 3149a70f27STrond Myklebust #include "internal.h" 3291d5b470SChuck Lever #include "iostat.h" 33def6ed7eSAndy Adamson #include "nfs4_fs.h" 34074cc1deSTrond Myklebust #include "fscache.h" 3594ad1c80SFred Isaman #include "pnfs.h" 361da177e4SLinus Torvalds 37f4ce1299STrond Myklebust #include "nfstrace.h" 38f4ce1299STrond Myklebust 391da177e4SLinus Torvalds #define NFSDBG_FACILITY NFSDBG_PAGECACHE 401da177e4SLinus Torvalds 411da177e4SLinus Torvalds #define MIN_POOL_WRITE (32) 421da177e4SLinus Torvalds #define MIN_POOL_COMMIT (4) 431da177e4SLinus Torvalds 44919e3bd9STrond Myklebust struct nfs_io_completion { 45919e3bd9STrond Myklebust void (*complete)(void *data); 46919e3bd9STrond Myklebust void *data; 47919e3bd9STrond Myklebust struct kref refcount; 48919e3bd9STrond Myklebust }; 49919e3bd9STrond Myklebust 501da177e4SLinus Torvalds /* 511da177e4SLinus Torvalds * Local function declarations 521da177e4SLinus Torvalds */ 53f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req); 54788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops; 55061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops; 56f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops; 574a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops; 58d4581383SWeston Andros Adamson static void nfs_clear_request_commit(struct nfs_page *req); 5902d1426cSWeston Andros Adamson static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 6002d1426cSWeston Andros Adamson struct inode *inode); 613a3908c8STrond Myklebust static struct nfs_page * 623a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 633a3908c8STrond Myklebust struct page *page); 641da177e4SLinus Torvalds 65e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep; 663feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool; 670b7c0153SFred Isaman static struct kmem_cache *nfs_cdata_cachep; 681da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool; 691da177e4SLinus Torvalds 70518662e0SNeilBrown struct nfs_commit_data *nfs_commitdata_alloc(bool never_fail) 711da177e4SLinus Torvalds { 72518662e0SNeilBrown struct nfs_commit_data *p; 7340859d7eSChuck Lever 74518662e0SNeilBrown if (never_fail) 75518662e0SNeilBrown p = mempool_alloc(nfs_commit_mempool, GFP_NOIO); 76518662e0SNeilBrown else { 77518662e0SNeilBrown /* It is OK to do some reclaim, not no safe to wait 78518662e0SNeilBrown * for anything to be returned to the pool. 79518662e0SNeilBrown * mempool_alloc() cannot handle that particular combination, 80518662e0SNeilBrown * so we need two separate attempts. 81518662e0SNeilBrown */ 82518662e0SNeilBrown p = mempool_alloc(nfs_commit_mempool, GFP_NOWAIT); 83518662e0SNeilBrown if (!p) 84518662e0SNeilBrown p = kmem_cache_alloc(nfs_cdata_cachep, GFP_NOIO | 85518662e0SNeilBrown __GFP_NOWARN | __GFP_NORETRY); 86518662e0SNeilBrown if (!p) 87518662e0SNeilBrown return NULL; 88518662e0SNeilBrown } 89518662e0SNeilBrown 901da177e4SLinus Torvalds memset(p, 0, sizeof(*p)); 911da177e4SLinus Torvalds INIT_LIST_HEAD(&p->pages); 921da177e4SLinus Torvalds return p; 931da177e4SLinus Torvalds } 94e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_alloc); 951da177e4SLinus Torvalds 960b7c0153SFred Isaman void nfs_commit_free(struct nfs_commit_data *p) 971da177e4SLinus Torvalds { 981da177e4SLinus Torvalds mempool_free(p, nfs_commit_mempool); 991da177e4SLinus Torvalds } 100e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commit_free); 1011da177e4SLinus Torvalds 1021e7f3a48SWeston Andros Adamson static struct nfs_pgio_header *nfs_writehdr_alloc(void) 1033feb2d49STrond Myklebust { 1041e7f3a48SWeston Andros Adamson struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO); 1053feb2d49STrond Myklebust 1063feb2d49STrond Myklebust memset(p, 0, sizeof(*p)); 107fbe77c30SBenjamin Coddington p->rw_mode = FMODE_WRITE; 1083feb2d49STrond Myklebust return p; 1093feb2d49STrond Myklebust } 1103feb2d49STrond Myklebust 1111e7f3a48SWeston Andros Adamson static void nfs_writehdr_free(struct nfs_pgio_header *hdr) 1126c75dc0dSFred Isaman { 1131e7f3a48SWeston Andros Adamson mempool_free(hdr, nfs_wdata_mempool); 1143feb2d49STrond Myklebust } 1151da177e4SLinus Torvalds 116919e3bd9STrond Myklebust static struct nfs_io_completion *nfs_io_completion_alloc(gfp_t gfp_flags) 117919e3bd9STrond Myklebust { 118919e3bd9STrond Myklebust return kmalloc(sizeof(struct nfs_io_completion), gfp_flags); 119919e3bd9STrond Myklebust } 120919e3bd9STrond Myklebust 121919e3bd9STrond Myklebust static void nfs_io_completion_init(struct nfs_io_completion *ioc, 122919e3bd9STrond Myklebust void (*complete)(void *), void *data) 123919e3bd9STrond Myklebust { 124919e3bd9STrond Myklebust ioc->complete = complete; 125919e3bd9STrond Myklebust ioc->data = data; 126919e3bd9STrond Myklebust kref_init(&ioc->refcount); 127919e3bd9STrond Myklebust } 128919e3bd9STrond Myklebust 129919e3bd9STrond Myklebust static void nfs_io_completion_release(struct kref *kref) 130919e3bd9STrond Myklebust { 131919e3bd9STrond Myklebust struct nfs_io_completion *ioc = container_of(kref, 132919e3bd9STrond Myklebust struct nfs_io_completion, refcount); 133919e3bd9STrond Myklebust ioc->complete(ioc->data); 134919e3bd9STrond Myklebust kfree(ioc); 135919e3bd9STrond Myklebust } 136919e3bd9STrond Myklebust 137919e3bd9STrond Myklebust static void nfs_io_completion_get(struct nfs_io_completion *ioc) 138919e3bd9STrond Myklebust { 139919e3bd9STrond Myklebust if (ioc != NULL) 140919e3bd9STrond Myklebust kref_get(&ioc->refcount); 141919e3bd9STrond Myklebust } 142919e3bd9STrond Myklebust 143919e3bd9STrond Myklebust static void nfs_io_completion_put(struct nfs_io_completion *ioc) 144919e3bd9STrond Myklebust { 145919e3bd9STrond Myklebust if (ioc != NULL) 146919e3bd9STrond Myklebust kref_put(&ioc->refcount, nfs_io_completion_release); 147919e3bd9STrond Myklebust } 148919e3bd9STrond Myklebust 149bd37d6fcSTrond Myklebust static struct nfs_page * 150bd37d6fcSTrond Myklebust nfs_page_private_request(struct page *page) 151bd37d6fcSTrond Myklebust { 152bd37d6fcSTrond Myklebust if (!PagePrivate(page)) 153bd37d6fcSTrond Myklebust return NULL; 154bd37d6fcSTrond Myklebust return (struct nfs_page *)page_private(page); 155bd37d6fcSTrond Myklebust } 156bd37d6fcSTrond Myklebust 15784d3a9a9SWeston Andros Adamson /* 15884d3a9a9SWeston Andros Adamson * nfs_page_find_head_request_locked - find head request associated with @page 15984d3a9a9SWeston Andros Adamson * 16084d3a9a9SWeston Andros Adamson * must be called while holding the inode lock. 16184d3a9a9SWeston Andros Adamson * 16284d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 16384d3a9a9SWeston Andros Adamson */ 16429418aa4SMel Gorman static struct nfs_page * 165b30d2f04STrond Myklebust nfs_page_find_private_request(struct page *page) 166277459d2STrond Myklebust { 1674b9bb25bSTrond Myklebust struct address_space *mapping = page_file_mapping(page); 168bd37d6fcSTrond Myklebust struct nfs_page *req; 169277459d2STrond Myklebust 170b30d2f04STrond Myklebust if (!PagePrivate(page)) 171b30d2f04STrond Myklebust return NULL; 1724b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 173bd37d6fcSTrond Myklebust req = nfs_page_private_request(page); 17484d3a9a9SWeston Andros Adamson if (req) { 17584d3a9a9SWeston Andros Adamson WARN_ON_ONCE(req->wb_head != req); 17629418aa4SMel Gorman kref_get(&req->wb_kref); 17784d3a9a9SWeston Andros Adamson } 1784b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 179b30d2f04STrond Myklebust return req; 180b30d2f04STrond Myklebust } 18129418aa4SMel Gorman 182b30d2f04STrond Myklebust static struct nfs_page * 183b30d2f04STrond Myklebust nfs_page_find_swap_request(struct page *page) 184b30d2f04STrond Myklebust { 185b30d2f04STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 186b30d2f04STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 187b30d2f04STrond Myklebust struct nfs_page *req = NULL; 188b30d2f04STrond Myklebust if (!PageSwapCache(page)) 189b30d2f04STrond Myklebust return NULL; 190e824f99aSTrond Myklebust mutex_lock(&nfsi->commit_mutex); 191b30d2f04STrond Myklebust if (PageSwapCache(page)) { 192b30d2f04STrond Myklebust req = nfs_page_search_commits_for_head_request_locked(nfsi, 193b30d2f04STrond Myklebust page); 194b30d2f04STrond Myklebust if (req) { 195b30d2f04STrond Myklebust WARN_ON_ONCE(req->wb_head != req); 196b30d2f04STrond Myklebust kref_get(&req->wb_kref); 197b30d2f04STrond Myklebust } 198b30d2f04STrond Myklebust } 199e824f99aSTrond Myklebust mutex_unlock(&nfsi->commit_mutex); 200277459d2STrond Myklebust return req; 201277459d2STrond Myklebust } 202277459d2STrond Myklebust 20384d3a9a9SWeston Andros Adamson /* 20484d3a9a9SWeston Andros Adamson * nfs_page_find_head_request - find head request associated with @page 20584d3a9a9SWeston Andros Adamson * 20684d3a9a9SWeston Andros Adamson * returns matching head request with reference held, or NULL if not found. 20784d3a9a9SWeston Andros Adamson */ 20884d3a9a9SWeston Andros Adamson static struct nfs_page *nfs_page_find_head_request(struct page *page) 209277459d2STrond Myklebust { 210b30d2f04STrond Myklebust struct nfs_page *req; 211277459d2STrond Myklebust 212b30d2f04STrond Myklebust req = nfs_page_find_private_request(page); 213b30d2f04STrond Myklebust if (!req) 214b30d2f04STrond Myklebust req = nfs_page_find_swap_request(page); 215277459d2STrond Myklebust return req; 216277459d2STrond Myklebust } 217277459d2STrond Myklebust 2181da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */ 2191da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count) 2201da177e4SLinus Torvalds { 221d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 222a3d01454STrond Myklebust loff_t end, i_size; 223a3d01454STrond Myklebust pgoff_t end_index; 2241da177e4SLinus Torvalds 225a3d01454STrond Myklebust spin_lock(&inode->i_lock); 226a3d01454STrond Myklebust i_size = i_size_read(inode); 22709cbfeafSKirill A. Shutemov end_index = (i_size - 1) >> PAGE_SHIFT; 2288cd79788SHuang Ying if (i_size > 0 && page_index(page) < end_index) 229a3d01454STrond Myklebust goto out; 230d56b4ddfSMel Gorman end = page_file_offset(page) + ((loff_t)offset+count); 2311da177e4SLinus Torvalds if (i_size >= end) 232a3d01454STrond Myklebust goto out; 2331da177e4SLinus Torvalds i_size_write(inode, end); 234f6cdfa6dSTrond Myklebust NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE; 235a3d01454STrond Myklebust nfs_inc_stats(inode, NFSIOS_EXTENDWRITE); 236a3d01454STrond Myklebust out: 237a3d01454STrond Myklebust spin_unlock(&inode->i_lock); 2381da177e4SLinus Torvalds } 2391da177e4SLinus Torvalds 240a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */ 241a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page) 242a301b777STrond Myklebust { 243d56b4ddfSMel Gorman nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page)); 244a301b777STrond Myklebust } 245a301b777STrond Myklebust 246d72ddcbaSWeston Andros Adamson /* 247d72ddcbaSWeston Andros Adamson * nfs_page_group_search_locked 248d72ddcbaSWeston Andros Adamson * @head - head request of page group 249d72ddcbaSWeston Andros Adamson * @page_offset - offset into page 250d72ddcbaSWeston Andros Adamson * 251d72ddcbaSWeston Andros Adamson * Search page group with head @head to find a request that contains the 252d72ddcbaSWeston Andros Adamson * page offset @page_offset. 253d72ddcbaSWeston Andros Adamson * 254d72ddcbaSWeston Andros Adamson * Returns a pointer to the first matching nfs request, or NULL if no 255d72ddcbaSWeston Andros Adamson * match is found. 256d72ddcbaSWeston Andros Adamson * 257d72ddcbaSWeston Andros Adamson * Must be called with the page group lock held 258d72ddcbaSWeston Andros Adamson */ 259d72ddcbaSWeston Andros Adamson static struct nfs_page * 260d72ddcbaSWeston Andros Adamson nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset) 261d72ddcbaSWeston Andros Adamson { 262d72ddcbaSWeston Andros Adamson struct nfs_page *req; 263d72ddcbaSWeston Andros Adamson 264d72ddcbaSWeston Andros Adamson req = head; 265d72ddcbaSWeston Andros Adamson do { 266d72ddcbaSWeston Andros Adamson if (page_offset >= req->wb_pgbase && 267d72ddcbaSWeston Andros Adamson page_offset < (req->wb_pgbase + req->wb_bytes)) 268d72ddcbaSWeston Andros Adamson return req; 269d72ddcbaSWeston Andros Adamson 270d72ddcbaSWeston Andros Adamson req = req->wb_this_page; 271d72ddcbaSWeston Andros Adamson } while (req != head); 272d72ddcbaSWeston Andros Adamson 273d72ddcbaSWeston Andros Adamson return NULL; 274d72ddcbaSWeston Andros Adamson } 275d72ddcbaSWeston Andros Adamson 276d72ddcbaSWeston Andros Adamson /* 277d72ddcbaSWeston Andros Adamson * nfs_page_group_covers_page 278d72ddcbaSWeston Andros Adamson * @head - head request of page group 279d72ddcbaSWeston Andros Adamson * 280d72ddcbaSWeston Andros Adamson * Return true if the page group with head @head covers the whole page, 281d72ddcbaSWeston Andros Adamson * returns false otherwise 282d72ddcbaSWeston Andros Adamson */ 283d72ddcbaSWeston Andros Adamson static bool nfs_page_group_covers_page(struct nfs_page *req) 284d72ddcbaSWeston Andros Adamson { 285d72ddcbaSWeston Andros Adamson struct nfs_page *tmp; 286d72ddcbaSWeston Andros Adamson unsigned int pos = 0; 287d72ddcbaSWeston Andros Adamson unsigned int len = nfs_page_length(req->wb_page); 288d72ddcbaSWeston Andros Adamson 2891344b7eaSTrond Myklebust nfs_page_group_lock(req); 290d72ddcbaSWeston Andros Adamson 2917e8a30f8STrond Myklebust for (;;) { 292d72ddcbaSWeston Andros Adamson tmp = nfs_page_group_search_locked(req->wb_head, pos); 2937e8a30f8STrond Myklebust if (!tmp) 2947e8a30f8STrond Myklebust break; 2957e8a30f8STrond Myklebust pos = tmp->wb_pgbase + tmp->wb_bytes; 296d72ddcbaSWeston Andros Adamson } 297d72ddcbaSWeston Andros Adamson 298d72ddcbaSWeston Andros Adamson nfs_page_group_unlock(req); 2997e8a30f8STrond Myklebust return pos >= len; 300d72ddcbaSWeston Andros Adamson } 301d72ddcbaSWeston Andros Adamson 3021da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request 3031da177e4SLinus Torvalds * covers the full page. 3041da177e4SLinus Torvalds */ 305d72ddcbaSWeston Andros Adamson static void nfs_mark_uptodate(struct nfs_page *req) 3061da177e4SLinus Torvalds { 307d72ddcbaSWeston Andros Adamson if (PageUptodate(req->wb_page)) 3081da177e4SLinus Torvalds return; 309d72ddcbaSWeston Andros Adamson if (!nfs_page_group_covers_page(req)) 3101da177e4SLinus Torvalds return; 311d72ddcbaSWeston Andros Adamson SetPageUptodate(req->wb_page); 3121da177e4SLinus Torvalds } 3131da177e4SLinus Torvalds 3141da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc) 3151da177e4SLinus Torvalds { 316e87b4c7aSNeilBrown int ret = 0; 317cca588d6STrond Myklebust 318e87b4c7aSNeilBrown if (wbc->sync_mode == WB_SYNC_ALL) 319e87b4c7aSNeilBrown ret = FLUSH_COND_STABLE; 320e87b4c7aSNeilBrown return ret; 3211da177e4SLinus Torvalds } 3221da177e4SLinus Torvalds 3231da177e4SLinus Torvalds /* 32489a09141SPeter Zijlstra * NFS congestion control 32589a09141SPeter Zijlstra */ 32689a09141SPeter Zijlstra 32789a09141SPeter Zijlstra int nfs_congestion_kb; 32889a09141SPeter Zijlstra 32989a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10)) 33089a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH \ 33189a09141SPeter Zijlstra (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2)) 33289a09141SPeter Zijlstra 333deed85e7STrond Myklebust static void nfs_set_page_writeback(struct page *page) 33489a09141SPeter Zijlstra { 3350db10944SJan Kara struct inode *inode = page_file_mapping(page)->host; 3360db10944SJan Kara struct nfs_server *nfss = NFS_SERVER(inode); 3375a6d41b3STrond Myklebust int ret = test_set_page_writeback(page); 3385a6d41b3STrond Myklebust 339deed85e7STrond Myklebust WARN_ON_ONCE(ret != 0); 34089a09141SPeter Zijlstra 341277866a0SPeter Zijlstra if (atomic_long_inc_return(&nfss->writeback) > 3420db10944SJan Kara NFS_CONGESTION_ON_THRESH) 3430db10944SJan Kara set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 34489a09141SPeter Zijlstra } 34589a09141SPeter Zijlstra 34620633f04SWeston Andros Adamson static void nfs_end_page_writeback(struct nfs_page *req) 34789a09141SPeter Zijlstra { 34820633f04SWeston Andros Adamson struct inode *inode = page_file_mapping(req->wb_page)->host; 34989a09141SPeter Zijlstra struct nfs_server *nfss = NFS_SERVER(inode); 35031a01f09STrond Myklebust bool is_done; 35189a09141SPeter Zijlstra 35231a01f09STrond Myklebust is_done = nfs_page_group_sync_on_bit(req, PG_WB_END); 35331a01f09STrond Myklebust nfs_unlock_request(req); 35431a01f09STrond Myklebust if (!is_done) 35520633f04SWeston Andros Adamson return; 35620633f04SWeston Andros Adamson 35720633f04SWeston Andros Adamson end_page_writeback(req->wb_page); 358c4dc4beeSPeter Zijlstra if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 3590db10944SJan Kara clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC); 36089a09141SPeter Zijlstra } 36189a09141SPeter Zijlstra 362d4581383SWeston Andros Adamson /* 363d4581383SWeston Andros Adamson * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req 364d4581383SWeston Andros Adamson * 365d4581383SWeston Andros Adamson * this is a helper function for nfs_lock_and_join_requests 366d4581383SWeston Andros Adamson * 367d4581383SWeston Andros Adamson * @inode - inode associated with request page group, must be holding inode lock 368d4581383SWeston Andros Adamson * @head - head request of page group, must be holding head lock 369d4581383SWeston Andros Adamson * @req - request that couldn't lock and needs to wait on the req bit lock 370d4581383SWeston Andros Adamson * 371b5bab9bfSTrond Myklebust * NOTE: this must be called holding page_group bit lock 372b5bab9bfSTrond Myklebust * which will be released before returning. 373d4581383SWeston Andros Adamson * 374d4581383SWeston Andros Adamson * returns 0 on success, < 0 on error. 375d4581383SWeston Andros Adamson */ 37674a6d4b5STrond Myklebust static void 37774a6d4b5STrond Myklebust nfs_unroll_locks(struct inode *inode, struct nfs_page *head, 3786d17d653STrond Myklebust struct nfs_page *req) 379d4581383SWeston Andros Adamson { 380d4581383SWeston Andros Adamson struct nfs_page *tmp; 381e261f51fSTrond Myklebust 382d4581383SWeston Andros Adamson /* relinquish all the locks successfully grabbed this run */ 3835b2b5187STrond Myklebust for (tmp = head->wb_this_page ; tmp != req; tmp = tmp->wb_this_page) { 3845b2b5187STrond Myklebust if (!kref_read(&tmp->wb_kref)) 3855b2b5187STrond Myklebust continue; 3865b2b5187STrond Myklebust nfs_unlock_and_release_request(tmp); 3875b2b5187STrond Myklebust } 388e261f51fSTrond Myklebust } 389d4581383SWeston Andros Adamson 390d4581383SWeston Andros Adamson /* 391d4581383SWeston Andros Adamson * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests 392d4581383SWeston Andros Adamson * 393d4581383SWeston Andros Adamson * @destroy_list - request list (using wb_this_page) terminated by @old_head 394d4581383SWeston Andros Adamson * @old_head - the old head of the list 395d4581383SWeston Andros Adamson * 396d4581383SWeston Andros Adamson * All subrequests must be locked and removed from all lists, so at this point 397d4581383SWeston Andros Adamson * they are only "active" in this function, and possibly in nfs_wait_on_request 398d4581383SWeston Andros Adamson * with a reference held by some other context. 399d4581383SWeston Andros Adamson */ 400d4581383SWeston Andros Adamson static void 401d4581383SWeston Andros Adamson nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list, 402b66aaa8dSTrond Myklebust struct nfs_page *old_head, 403b66aaa8dSTrond Myklebust struct inode *inode) 404d4581383SWeston Andros Adamson { 405d4581383SWeston Andros Adamson while (destroy_list) { 406d4581383SWeston Andros Adamson struct nfs_page *subreq = destroy_list; 407d4581383SWeston Andros Adamson 408d4581383SWeston Andros Adamson destroy_list = (subreq->wb_this_page == old_head) ? 409d4581383SWeston Andros Adamson NULL : subreq->wb_this_page; 410d4581383SWeston Andros Adamson 411d4581383SWeston Andros Adamson WARN_ON_ONCE(old_head != subreq->wb_head); 412d4581383SWeston Andros Adamson 413d4581383SWeston Andros Adamson /* make sure old group is not used */ 414d4581383SWeston Andros Adamson subreq->wb_this_page = subreq; 415d4581383SWeston Andros Adamson 416902a4c00STrond Myklebust clear_bit(PG_REMOVE, &subreq->wb_flags); 417902a4c00STrond Myklebust 4185b2b5187STrond Myklebust /* Note: races with nfs_page_group_destroy() */ 4195b2b5187STrond Myklebust if (!kref_read(&subreq->wb_kref)) { 4205b2b5187STrond Myklebust /* Check if we raced with nfs_page_group_destroy() */ 421902a4c00STrond Myklebust if (test_and_clear_bit(PG_TEARDOWN, &subreq->wb_flags)) 4225b2b5187STrond Myklebust nfs_free_request(subreq); 4235b2b5187STrond Myklebust continue; 4245b2b5187STrond Myklebust } 425d4581383SWeston Andros Adamson 4265b2b5187STrond Myklebust subreq->wb_head = subreq; 4275b2b5187STrond Myklebust 428b66aaa8dSTrond Myklebust if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) { 429d4581383SWeston Andros Adamson nfs_release_request(subreq); 430a6b6d5b8STrond Myklebust atomic_long_dec(&NFS_I(inode)->nrequests); 431d4581383SWeston Andros Adamson } 4325b2b5187STrond Myklebust 4335b2b5187STrond Myklebust /* subreq is now totally disconnected from page group or any 4345b2b5187STrond Myklebust * write / commit lists. last chance to wake any waiters */ 4355b2b5187STrond Myklebust nfs_unlock_and_release_request(subreq); 436d4581383SWeston Andros Adamson } 437d4581383SWeston Andros Adamson } 438d4581383SWeston Andros Adamson 439d4581383SWeston Andros Adamson /* 440d4581383SWeston Andros Adamson * nfs_lock_and_join_requests - join all subreqs to the head req and return 441d4581383SWeston Andros Adamson * a locked reference, cancelling any pending 442d4581383SWeston Andros Adamson * operations for this page. 443d4581383SWeston Andros Adamson * 444d4581383SWeston Andros Adamson * @page - the page used to lookup the "page group" of nfs_page structures 445d4581383SWeston Andros Adamson * 446d4581383SWeston Andros Adamson * This function joins all sub requests to the head request by first 447d4581383SWeston Andros Adamson * locking all requests in the group, cancelling any pending operations 448d4581383SWeston Andros Adamson * and finally updating the head request to cover the whole range covered by 449d4581383SWeston Andros Adamson * the (former) group. All subrequests are removed from any write or commit 450d4581383SWeston Andros Adamson * lists, unlinked from the group and destroyed. 451d4581383SWeston Andros Adamson * 452d4581383SWeston Andros Adamson * Returns a locked, referenced pointer to the head request - which after 453d4581383SWeston Andros Adamson * this call is guaranteed to be the only request associated with the page. 454d4581383SWeston Andros Adamson * Returns NULL if no requests are found for @page, or a ERR_PTR if an 455d4581383SWeston Andros Adamson * error was encountered. 456d4581383SWeston Andros Adamson */ 457d4581383SWeston Andros Adamson static struct nfs_page * 4586d17d653STrond Myklebust nfs_lock_and_join_requests(struct page *page) 459d4581383SWeston Andros Adamson { 460d4581383SWeston Andros Adamson struct inode *inode = page_file_mapping(page)->host; 461d4581383SWeston Andros Adamson struct nfs_page *head, *subreq; 462d4581383SWeston Andros Adamson struct nfs_page *destroy_list = NULL; 463d4581383SWeston Andros Adamson unsigned int total_bytes; 464d4581383SWeston Andros Adamson int ret; 465d4581383SWeston Andros Adamson 466d4581383SWeston Andros Adamson try_again: 467d4581383SWeston Andros Adamson /* 468d4581383SWeston Andros Adamson * A reference is taken only on the head request which acts as a 469d4581383SWeston Andros Adamson * reference to the whole page group - the group will not be destroyed 470d4581383SWeston Andros Adamson * until the head reference is released. 471d4581383SWeston Andros Adamson */ 472bd37d6fcSTrond Myklebust head = nfs_page_find_head_request(page); 473bd37d6fcSTrond Myklebust if (!head) 474d4581383SWeston Andros Adamson return NULL; 475d4581383SWeston Andros Adamson 476a0e265bcSTrond Myklebust /* lock the page head first in order to avoid an ABBA inefficiency */ 477a0e265bcSTrond Myklebust if (!nfs_lock_request(head)) { 478a0e265bcSTrond Myklebust ret = nfs_wait_on_request(head); 479a0e265bcSTrond Myklebust nfs_release_request(head); 480a0e265bcSTrond Myklebust if (ret < 0) 481a0e265bcSTrond Myklebust return ERR_PTR(ret); 482a0e265bcSTrond Myklebust goto try_again; 483a0e265bcSTrond Myklebust } 484bd37d6fcSTrond Myklebust 485bd37d6fcSTrond Myklebust /* Ensure that nobody removed the request before we locked it */ 486bd37d6fcSTrond Myklebust if (head != nfs_page_private_request(page) && !PageSwapCache(page)) { 487bd37d6fcSTrond Myklebust nfs_unlock_and_release_request(head); 488bd37d6fcSTrond Myklebust goto try_again; 489bd37d6fcSTrond Myklebust } 4907c3af975SWeston Andros Adamson 4911344b7eaSTrond Myklebust ret = nfs_page_group_lock(head); 4920671d8f1SMarkus Elfring if (ret < 0) 4930671d8f1SMarkus Elfring goto release_request; 4947c3af975SWeston Andros Adamson 4957c3af975SWeston Andros Adamson /* lock each request in the page group */ 496a0e265bcSTrond Myklebust total_bytes = head->wb_bytes; 497a0e265bcSTrond Myklebust for (subreq = head->wb_this_page; subreq != head; 498a0e265bcSTrond Myklebust subreq = subreq->wb_this_page) { 49974a6d4b5STrond Myklebust 5001bd5d6d0STrond Myklebust if (!kref_get_unless_zero(&subreq->wb_kref)) { 5011bd5d6d0STrond Myklebust if (subreq->wb_offset == head->wb_offset + total_bytes) 5021bd5d6d0STrond Myklebust total_bytes += subreq->wb_bytes; 5035b2b5187STrond Myklebust continue; 5041bd5d6d0STrond Myklebust } 5051bd5d6d0STrond Myklebust 50674a6d4b5STrond Myklebust while (!nfs_lock_request(subreq)) { 507b5bab9bfSTrond Myklebust /* 50874a6d4b5STrond Myklebust * Unlock page to allow nfs_page_group_sync_on_bit() 50974a6d4b5STrond Myklebust * to succeed 510b5bab9bfSTrond Myklebust */ 51174a6d4b5STrond Myklebust nfs_page_group_unlock(head); 51274a6d4b5STrond Myklebust ret = nfs_wait_on_request(subreq); 51374a6d4b5STrond Myklebust if (!ret) 5141344b7eaSTrond Myklebust ret = nfs_page_group_lock(head); 51574a6d4b5STrond Myklebust if (ret < 0) { 51674a6d4b5STrond Myklebust nfs_unroll_locks(inode, head, subreq); 5175b2b5187STrond Myklebust nfs_release_request(subreq); 5180671d8f1SMarkus Elfring goto release_request; 519d4581383SWeston Andros Adamson } 52074a6d4b5STrond Myklebust } 521e14bebf6STrond Myklebust /* 522e14bebf6STrond Myklebust * Subrequests are always contiguous, non overlapping 523e14bebf6STrond Myklebust * and in order - but may be repeated (mirrored writes). 524e14bebf6STrond Myklebust */ 525e14bebf6STrond Myklebust if (subreq->wb_offset == (head->wb_offset + total_bytes)) { 526e14bebf6STrond Myklebust /* keep track of how many bytes this group covers */ 527e14bebf6STrond Myklebust total_bytes += subreq->wb_bytes; 528e14bebf6STrond Myklebust } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset || 529e14bebf6STrond Myklebust ((subreq->wb_offset + subreq->wb_bytes) > 530e14bebf6STrond Myklebust (head->wb_offset + total_bytes)))) { 5318b77484fSTrond Myklebust nfs_page_group_unlock(head); 53274a6d4b5STrond Myklebust nfs_unroll_locks(inode, head, subreq); 5335b2b5187STrond Myklebust nfs_unlock_and_release_request(subreq); 5340671d8f1SMarkus Elfring ret = -EIO; 5350671d8f1SMarkus Elfring goto release_request; 536e14bebf6STrond Myklebust } 537a0e265bcSTrond Myklebust } 538d4581383SWeston Andros Adamson 539d4581383SWeston Andros Adamson /* Now that all requests are locked, make sure they aren't on any list. 540d4581383SWeston Andros Adamson * Commit list removal accounting is done after locks are dropped */ 541d4581383SWeston Andros Adamson subreq = head; 542d4581383SWeston Andros Adamson do { 543411a99adSWeston Andros Adamson nfs_clear_request_commit(subreq); 544d4581383SWeston Andros Adamson subreq = subreq->wb_this_page; 545d4581383SWeston Andros Adamson } while (subreq != head); 546d4581383SWeston Andros Adamson 547d4581383SWeston Andros Adamson /* unlink subrequests from head, destroy them later */ 548d4581383SWeston Andros Adamson if (head->wb_this_page != head) { 549d4581383SWeston Andros Adamson /* destroy list will be terminated by head */ 550d4581383SWeston Andros Adamson destroy_list = head->wb_this_page; 551d4581383SWeston Andros Adamson head->wb_this_page = head; 552d4581383SWeston Andros Adamson 553d4581383SWeston Andros Adamson /* change head request to cover whole range that 554d4581383SWeston Andros Adamson * the former page group covered */ 555d4581383SWeston Andros Adamson head->wb_bytes = total_bytes; 556d4581383SWeston Andros Adamson } 557d4581383SWeston Andros Adamson 558b66aaa8dSTrond Myklebust /* Postpone destruction of this request */ 559b66aaa8dSTrond Myklebust if (test_and_clear_bit(PG_REMOVE, &head->wb_flags)) { 560b66aaa8dSTrond Myklebust set_bit(PG_INODE_REF, &head->wb_flags); 561b66aaa8dSTrond Myklebust kref_get(&head->wb_kref); 562a6b6d5b8STrond Myklebust atomic_long_inc(&NFS_I(inode)->nrequests); 563b66aaa8dSTrond Myklebust } 564b66aaa8dSTrond Myklebust 565d4581383SWeston Andros Adamson nfs_page_group_unlock(head); 566d4581383SWeston Andros Adamson 567b66aaa8dSTrond Myklebust nfs_destroy_unlinked_subrequests(destroy_list, head, inode); 568d4581383SWeston Andros Adamson 569b5bab9bfSTrond Myklebust /* Did we lose a race with nfs_inode_remove_request()? */ 570b5bab9bfSTrond Myklebust if (!(PagePrivate(page) || PageSwapCache(page))) { 571b5bab9bfSTrond Myklebust nfs_unlock_and_release_request(head); 572b5bab9bfSTrond Myklebust return NULL; 573b5bab9bfSTrond Myklebust } 574b5bab9bfSTrond Myklebust 575bd37d6fcSTrond Myklebust /* still holds ref on head from nfs_page_find_head_request 576d4581383SWeston Andros Adamson * and still has lock on head from lock loop */ 577d4581383SWeston Andros Adamson return head; 5780671d8f1SMarkus Elfring 5790671d8f1SMarkus Elfring release_request: 5800671d8f1SMarkus Elfring nfs_unlock_and_release_request(head); 5810671d8f1SMarkus Elfring return ERR_PTR(ret); 582612c9384STrond Myklebust } 583074cc1deSTrond Myklebust 5840bcbf039SPeng Tao static void nfs_write_error_remove_page(struct nfs_page *req) 5850bcbf039SPeng Tao { 5860bcbf039SPeng Tao nfs_end_page_writeback(req); 5870bcbf039SPeng Tao generic_error_remove_page(page_file_mapping(req->wb_page), 5880bcbf039SPeng Tao req->wb_page); 5891f84ccdfSFred Isaman nfs_release_request(req); 5900bcbf039SPeng Tao } 5910bcbf039SPeng Tao 592a6598813STrond Myklebust static bool 593a6598813STrond Myklebust nfs_error_is_fatal_on_server(int err) 594a6598813STrond Myklebust { 595a6598813STrond Myklebust switch (err) { 596a6598813STrond Myklebust case 0: 597a6598813STrond Myklebust case -ERESTARTSYS: 598a6598813STrond Myklebust case -EINTR: 599a6598813STrond Myklebust return false; 600a6598813STrond Myklebust } 601a6598813STrond Myklebust return nfs_error_is_fatal(err); 602e261f51fSTrond Myklebust } 603074cc1deSTrond Myklebust 604074cc1deSTrond Myklebust /* 605074cc1deSTrond Myklebust * Find an associated nfs write request, and prepare to flush it out 606074cc1deSTrond Myklebust * May return an error if the user signalled nfs_wait_on_request(). 607074cc1deSTrond Myklebust */ 608074cc1deSTrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio, 6096d17d653STrond Myklebust struct page *page) 610074cc1deSTrond Myklebust { 611074cc1deSTrond Myklebust struct nfs_page *req; 612074cc1deSTrond Myklebust int ret = 0; 613074cc1deSTrond Myklebust 6146d17d653STrond Myklebust req = nfs_lock_and_join_requests(page); 615074cc1deSTrond Myklebust if (!req) 616074cc1deSTrond Myklebust goto out; 617074cc1deSTrond Myklebust ret = PTR_ERR(req); 618074cc1deSTrond Myklebust if (IS_ERR(req)) 619074cc1deSTrond Myklebust goto out; 620074cc1deSTrond Myklebust 621deed85e7STrond Myklebust nfs_set_page_writeback(page); 622deed85e7STrond Myklebust WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags)); 623074cc1deSTrond Myklebust 624deed85e7STrond Myklebust ret = 0; 625a6598813STrond Myklebust /* If there is a fatal error that covers this write, just exit */ 626a6598813STrond Myklebust if (nfs_error_is_fatal_on_server(req->wb_context->error)) 627a6598813STrond Myklebust goto out_launder; 628a6598813STrond Myklebust 629f8512ad0SFred Isaman if (!nfs_pageio_add_request(pgio, req)) { 630074cc1deSTrond Myklebust ret = pgio->pg_error; 6310bcbf039SPeng Tao /* 632c373fff7STrond Myklebust * Remove the problematic req upon fatal errors on the server 6330bcbf039SPeng Tao */ 6340bcbf039SPeng Tao if (nfs_error_is_fatal(ret)) { 6350bcbf039SPeng Tao nfs_context_set_write_error(req->wb_context, ret); 636c373fff7STrond Myklebust if (nfs_error_is_fatal_on_server(ret)) 637a6598813STrond Myklebust goto out_launder; 638d6c843b9SPeng Tao } 6390bcbf039SPeng Tao nfs_redirty_request(req); 6400bcbf039SPeng Tao ret = -EAGAIN; 64140f90271STrond Myklebust } else 64240f90271STrond Myklebust nfs_add_stats(page_file_mapping(page)->host, 64340f90271STrond Myklebust NFSIOS_WRITEPAGES, 1); 644074cc1deSTrond Myklebust out: 645074cc1deSTrond Myklebust return ret; 646a6598813STrond Myklebust out_launder: 647a6598813STrond Myklebust nfs_write_error_remove_page(req); 648a6598813STrond Myklebust return ret; 649e261f51fSTrond Myklebust } 650e261f51fSTrond Myklebust 651d6c843b9SPeng Tao static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, 652c373fff7STrond Myklebust struct nfs_pageio_descriptor *pgio) 653f758c885STrond Myklebust { 654cfb506e1STrond Myklebust int ret; 655f758c885STrond Myklebust 6568cd79788SHuang Ying nfs_pageio_cond_complete(pgio, page_index(page)); 6576d17d653STrond Myklebust ret = nfs_page_async_flush(pgio, page); 658cfb506e1STrond Myklebust if (ret == -EAGAIN) { 659cfb506e1STrond Myklebust redirty_page_for_writepage(wbc, page); 660cfb506e1STrond Myklebust ret = 0; 661cfb506e1STrond Myklebust } 662cfb506e1STrond Myklebust return ret; 663f758c885STrond Myklebust } 664f758c885STrond Myklebust 665e261f51fSTrond Myklebust /* 6661da177e4SLinus Torvalds * Write an mmapped page to the server. 6671da177e4SLinus Torvalds */ 668d6c843b9SPeng Tao static int nfs_writepage_locked(struct page *page, 669c373fff7STrond Myklebust struct writeback_control *wbc) 6701da177e4SLinus Torvalds { 671f758c885STrond Myklebust struct nfs_pageio_descriptor pgio; 67240f90271STrond Myklebust struct inode *inode = page_file_mapping(page)->host; 673e261f51fSTrond Myklebust int err; 6741da177e4SLinus Torvalds 67540f90271STrond Myklebust nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE); 676811ed92eSTrond Myklebust nfs_pageio_init_write(&pgio, inode, 0, 677a20c93e3SChristoph Hellwig false, &nfs_async_write_completion_ops); 678c373fff7STrond Myklebust err = nfs_do_writepage(page, wbc, &pgio); 679f758c885STrond Myklebust nfs_pageio_complete(&pgio); 680f758c885STrond Myklebust if (err < 0) 6814d770ccfSTrond Myklebust return err; 682f758c885STrond Myklebust if (pgio.pg_error < 0) 683f758c885STrond Myklebust return pgio.pg_error; 684f758c885STrond Myklebust return 0; 6854d770ccfSTrond Myklebust } 6864d770ccfSTrond Myklebust 6874d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc) 6884d770ccfSTrond Myklebust { 689f758c885STrond Myklebust int ret; 6904d770ccfSTrond Myklebust 691c373fff7STrond Myklebust ret = nfs_writepage_locked(page, wbc); 6921da177e4SLinus Torvalds unlock_page(page); 693f758c885STrond Myklebust return ret; 694f758c885STrond Myklebust } 695f758c885STrond Myklebust 696f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data) 697f758c885STrond Myklebust { 698f758c885STrond Myklebust int ret; 699f758c885STrond Myklebust 700c373fff7STrond Myklebust ret = nfs_do_writepage(page, wbc, data); 701f758c885STrond Myklebust unlock_page(page); 702f758c885STrond Myklebust return ret; 7031da177e4SLinus Torvalds } 7041da177e4SLinus Torvalds 705919e3bd9STrond Myklebust static void nfs_io_completion_commit(void *inode) 706919e3bd9STrond Myklebust { 707919e3bd9STrond Myklebust nfs_commit_inode(inode, 0); 708919e3bd9STrond Myklebust } 709919e3bd9STrond Myklebust 7101da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc) 7111da177e4SLinus Torvalds { 7121da177e4SLinus Torvalds struct inode *inode = mapping->host; 713c63c7b05STrond Myklebust struct nfs_pageio_descriptor pgio; 714919e3bd9STrond Myklebust struct nfs_io_completion *ioc = nfs_io_completion_alloc(GFP_NOFS); 7151da177e4SLinus Torvalds int err; 7161da177e4SLinus Torvalds 71791d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES); 71891d5b470SChuck Lever 719919e3bd9STrond Myklebust if (ioc) 720919e3bd9STrond Myklebust nfs_io_completion_init(ioc, nfs_io_completion_commit, inode); 721919e3bd9STrond Myklebust 722a20c93e3SChristoph Hellwig nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false, 723a20c93e3SChristoph Hellwig &nfs_async_write_completion_ops); 724919e3bd9STrond Myklebust pgio.pg_io_completion = ioc; 725f758c885STrond Myklebust err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio); 726c63c7b05STrond Myklebust nfs_pageio_complete(&pgio); 727919e3bd9STrond Myklebust nfs_io_completion_put(ioc); 72872cb77f4STrond Myklebust 729f758c885STrond Myklebust if (err < 0) 73072cb77f4STrond Myklebust goto out_err; 73172cb77f4STrond Myklebust err = pgio.pg_error; 73272cb77f4STrond Myklebust if (err < 0) 73372cb77f4STrond Myklebust goto out_err; 734c63c7b05STrond Myklebust return 0; 73572cb77f4STrond Myklebust out_err: 73672cb77f4STrond Myklebust return err; 7371da177e4SLinus Torvalds } 7381da177e4SLinus Torvalds 7391da177e4SLinus Torvalds /* 7401da177e4SLinus Torvalds * Insert a write request into an inode 7411da177e4SLinus Torvalds */ 742d6d6dc7cSFred Isaman static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req) 7431da177e4SLinus Torvalds { 7444b9bb25bSTrond Myklebust struct address_space *mapping = page_file_mapping(req->wb_page); 7451da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 746e7d39069STrond Myklebust 7472bfc6e56SWeston Andros Adamson WARN_ON_ONCE(req->wb_this_page != req); 7482bfc6e56SWeston Andros Adamson 749e7d39069STrond Myklebust /* Lock the request! */ 7507ad84aa9STrond Myklebust nfs_lock_request(req); 751e7d39069STrond Myklebust 75229418aa4SMel Gorman /* 75329418aa4SMel Gorman * Swap-space should not get truncated. Hence no need to plug the race 75429418aa4SMel Gorman * with invalidate/truncate. 75529418aa4SMel Gorman */ 7564b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 7574b9bb25bSTrond Myklebust if (!nfs_have_writebacks(inode) && 7581eb5d98fSJeff Layton NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 7591eb5d98fSJeff Layton inode_inc_iversion_raw(inode); 76029418aa4SMel Gorman if (likely(!PageSwapCache(req->wb_page))) { 7612df485a7STrond Myklebust set_bit(PG_MAPPED, &req->wb_flags); 762deb7d638STrond Myklebust SetPagePrivate(req->wb_page); 763277459d2STrond Myklebust set_page_private(req->wb_page, (unsigned long)req); 76429418aa4SMel Gorman } 7654b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 766a6b6d5b8STrond Myklebust atomic_long_inc(&nfsi->nrequests); 76717089a29SWeston Andros Adamson /* this a head request for a page group - mark it as having an 768cb1410c7SWeston Andros Adamson * extra reference so sub groups can follow suit. 769cb1410c7SWeston Andros Adamson * This flag also informs pgio layer when to bump nrequests when 770cb1410c7SWeston Andros Adamson * adding subrequests. */ 77117089a29SWeston Andros Adamson WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags)); 772c03b4024STrond Myklebust kref_get(&req->wb_kref); 7731da177e4SLinus Torvalds } 7741da177e4SLinus Torvalds 7751da177e4SLinus Torvalds /* 77689a09141SPeter Zijlstra * Remove a write request from an inode 7771da177e4SLinus Torvalds */ 7781da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req) 7791da177e4SLinus Torvalds { 7804b9bb25bSTrond Myklebust struct address_space *mapping = page_file_mapping(req->wb_page); 7814b9bb25bSTrond Myklebust struct inode *inode = mapping->host; 7821da177e4SLinus Torvalds struct nfs_inode *nfsi = NFS_I(inode); 78320633f04SWeston Andros Adamson struct nfs_page *head; 78420633f04SWeston Andros Adamson 785a6b6d5b8STrond Myklebust atomic_long_dec(&nfsi->nrequests); 78620633f04SWeston Andros Adamson if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) { 78720633f04SWeston Andros Adamson head = req->wb_head; 7881da177e4SLinus Torvalds 7894b9bb25bSTrond Myklebust spin_lock(&mapping->private_lock); 79067911c8fSAnna Schumaker if (likely(head->wb_page && !PageSwapCache(head->wb_page))) { 79120633f04SWeston Andros Adamson set_page_private(head->wb_page, 0); 79220633f04SWeston Andros Adamson ClearPagePrivate(head->wb_page); 79320633f04SWeston Andros Adamson clear_bit(PG_MAPPED, &head->wb_flags); 79429418aa4SMel Gorman } 7954b9bb25bSTrond Myklebust spin_unlock(&mapping->private_lock); 79620633f04SWeston Andros Adamson } 79717089a29SWeston Andros Adamson 79817089a29SWeston Andros Adamson if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) 7991da177e4SLinus Torvalds nfs_release_request(req); 8001da177e4SLinus Torvalds } 8011da177e4SLinus Torvalds 80261822ab5STrond Myklebust static void 8036d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req) 80461822ab5STrond Myklebust { 80567911c8fSAnna Schumaker if (req->wb_page) 80661822ab5STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 80761822ab5STrond Myklebust } 80861822ab5STrond Myklebust 8093a3908c8STrond Myklebust /* 8103a3908c8STrond Myklebust * nfs_page_search_commits_for_head_request_locked 8113a3908c8STrond Myklebust * 8123a3908c8STrond Myklebust * Search through commit lists on @inode for the head request for @page. 8133a3908c8STrond Myklebust * Must be called while holding the inode (which is cinfo) lock. 8143a3908c8STrond Myklebust * 8153a3908c8STrond Myklebust * Returns the head request if found, or NULL if not found. 8163a3908c8STrond Myklebust */ 8173a3908c8STrond Myklebust static struct nfs_page * 8183a3908c8STrond Myklebust nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi, 8193a3908c8STrond Myklebust struct page *page) 8203a3908c8STrond Myklebust { 8213a3908c8STrond Myklebust struct nfs_page *freq, *t; 8223a3908c8STrond Myklebust struct nfs_commit_info cinfo; 8233a3908c8STrond Myklebust struct inode *inode = &nfsi->vfs_inode; 8243a3908c8STrond Myklebust 8253a3908c8STrond Myklebust nfs_init_cinfo_from_inode(&cinfo, inode); 8263a3908c8STrond Myklebust 8273a3908c8STrond Myklebust /* search through pnfs commit lists */ 8283a3908c8STrond Myklebust freq = pnfs_search_commit_reqs(inode, &cinfo, page); 8293a3908c8STrond Myklebust if (freq) 8303a3908c8STrond Myklebust return freq->wb_head; 8313a3908c8STrond Myklebust 8323a3908c8STrond Myklebust /* Linearly search the commit list for the correct request */ 8333a3908c8STrond Myklebust list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) { 8343a3908c8STrond Myklebust if (freq->wb_page == page) 8353a3908c8STrond Myklebust return freq->wb_head; 8363a3908c8STrond Myklebust } 8373a3908c8STrond Myklebust 8383a3908c8STrond Myklebust return NULL; 8393a3908c8STrond Myklebust } 8403a3908c8STrond Myklebust 8418dd37758STrond Myklebust /** 84286d80f97STrond Myklebust * nfs_request_add_commit_list_locked - add request to a commit list 84386d80f97STrond Myklebust * @req: pointer to a struct nfs_page 84486d80f97STrond Myklebust * @dst: commit list head 84586d80f97STrond Myklebust * @cinfo: holds list lock and accounting info 84686d80f97STrond Myklebust * 84786d80f97STrond Myklebust * This sets the PG_CLEAN bit, updates the cinfo count of 84886d80f97STrond Myklebust * number of outstanding requests requiring a commit as well as 84986d80f97STrond Myklebust * the MM page stats. 85086d80f97STrond Myklebust * 851e824f99aSTrond Myklebust * The caller must hold NFS_I(cinfo->inode)->commit_mutex, and the 852e824f99aSTrond Myklebust * nfs_page lock. 85386d80f97STrond Myklebust */ 85486d80f97STrond Myklebust void 85586d80f97STrond Myklebust nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst, 85686d80f97STrond Myklebust struct nfs_commit_info *cinfo) 85786d80f97STrond Myklebust { 85886d80f97STrond Myklebust set_bit(PG_CLEAN, &req->wb_flags); 85986d80f97STrond Myklebust nfs_list_add_request(req, dst); 8605cb953d4STrond Myklebust atomic_long_inc(&cinfo->mds->ncommit); 86186d80f97STrond Myklebust } 86286d80f97STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked); 86386d80f97STrond Myklebust 86486d80f97STrond Myklebust /** 8658dd37758STrond Myklebust * nfs_request_add_commit_list - add request to a commit list 8668dd37758STrond Myklebust * @req: pointer to a struct nfs_page 867ea2cf228SFred Isaman * @dst: commit list head 868ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8698dd37758STrond Myklebust * 870ea2cf228SFred Isaman * This sets the PG_CLEAN bit, updates the cinfo count of 8718dd37758STrond Myklebust * number of outstanding requests requiring a commit as well as 8728dd37758STrond Myklebust * the MM page stats. 8738dd37758STrond Myklebust * 874ea2cf228SFred Isaman * The caller must _not_ hold the cinfo->lock, but must be 8758dd37758STrond Myklebust * holding the nfs_page lock. 8768dd37758STrond Myklebust */ 8778dd37758STrond Myklebust void 8786272dcc6SAnna Schumaker nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo) 8798dd37758STrond Myklebust { 880e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 8816272dcc6SAnna Schumaker nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo); 882e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 88367911c8fSAnna Schumaker if (req->wb_page) 88486d80f97STrond Myklebust nfs_mark_page_unstable(req->wb_page, cinfo); 8858dd37758STrond Myklebust } 8868dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_add_commit_list); 8878dd37758STrond Myklebust 8888dd37758STrond Myklebust /** 8898dd37758STrond Myklebust * nfs_request_remove_commit_list - Remove request from a commit list 8908dd37758STrond Myklebust * @req: pointer to a nfs_page 891ea2cf228SFred Isaman * @cinfo: holds list lock and accounting info 8928dd37758STrond Myklebust * 893ea2cf228SFred Isaman * This clears the PG_CLEAN bit, and updates the cinfo's count of 8948dd37758STrond Myklebust * number of outstanding requests requiring a commit 8958dd37758STrond Myklebust * It does not update the MM page stats. 8968dd37758STrond Myklebust * 897ea2cf228SFred Isaman * The caller _must_ hold the cinfo->lock and the nfs_page lock. 8988dd37758STrond Myklebust */ 8998dd37758STrond Myklebust void 900ea2cf228SFred Isaman nfs_request_remove_commit_list(struct nfs_page *req, 901ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 9028dd37758STrond Myklebust { 9038dd37758STrond Myklebust if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) 9048dd37758STrond Myklebust return; 9058dd37758STrond Myklebust nfs_list_remove_request(req); 9065cb953d4STrond Myklebust atomic_long_dec(&cinfo->mds->ncommit); 9078dd37758STrond Myklebust } 9088dd37758STrond Myklebust EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list); 9098dd37758STrond Myklebust 910ea2cf228SFred Isaman static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo, 911ea2cf228SFred Isaman struct inode *inode) 912ea2cf228SFred Isaman { 913fe238e60SDave Wysochanski cinfo->inode = inode; 914ea2cf228SFred Isaman cinfo->mds = &NFS_I(inode)->commit_info; 915ea2cf228SFred Isaman cinfo->ds = pnfs_get_ds_info(inode); 916b359f9d0SFred Isaman cinfo->dreq = NULL; 917f453a54aSFred Isaman cinfo->completion_ops = &nfs_commit_completion_ops; 918ea2cf228SFred Isaman } 919ea2cf228SFred Isaman 920ea2cf228SFred Isaman void nfs_init_cinfo(struct nfs_commit_info *cinfo, 921ea2cf228SFred Isaman struct inode *inode, 922ea2cf228SFred Isaman struct nfs_direct_req *dreq) 923ea2cf228SFred Isaman { 9241763da12SFred Isaman if (dreq) 9251763da12SFred Isaman nfs_init_cinfo_from_dreq(cinfo, dreq); 9261763da12SFred Isaman else 927ea2cf228SFred Isaman nfs_init_cinfo_from_inode(cinfo, inode); 928ea2cf228SFred Isaman } 929ea2cf228SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_cinfo); 9308dd37758STrond Myklebust 9311da177e4SLinus Torvalds /* 9321da177e4SLinus Torvalds * Add a request to the inode's commit list. 9331da177e4SLinus Torvalds */ 9341763da12SFred Isaman void 935ea2cf228SFred Isaman nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg, 936b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, u32 ds_commit_idx) 9371da177e4SLinus Torvalds { 938b57ff130SWeston Andros Adamson if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx)) 9398dd37758STrond Myklebust return; 9406272dcc6SAnna Schumaker nfs_request_add_commit_list(req, cinfo); 9411da177e4SLinus Torvalds } 9428e821cadSTrond Myklebust 943d6d6dc7cSFred Isaman static void 944d6d6dc7cSFred Isaman nfs_clear_page_commit(struct page *page) 945e468bae9STrond Myklebust { 94611fb9989SMel Gorman dec_node_page_state(page, NR_UNSTABLE_NFS); 94793f78d88STejun Heo dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb, 94893f78d88STejun Heo WB_RECLAIMABLE); 949e468bae9STrond Myklebust } 950d6d6dc7cSFred Isaman 951b5bab9bfSTrond Myklebust /* Called holding the request lock on @req */ 9528dd37758STrond Myklebust static void 953d6d6dc7cSFred Isaman nfs_clear_request_commit(struct nfs_page *req) 954d6d6dc7cSFred Isaman { 9558dd37758STrond Myklebust if (test_bit(PG_CLEAN, &req->wb_flags)) { 9562b0143b5SDavid Howells struct inode *inode = d_inode(req->wb_context->dentry); 957ea2cf228SFred Isaman struct nfs_commit_info cinfo; 958d6d6dc7cSFred Isaman 959ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 960e824f99aSTrond Myklebust mutex_lock(&NFS_I(inode)->commit_mutex); 961ea2cf228SFred Isaman if (!pnfs_clear_request_commit(req, &cinfo)) { 962ea2cf228SFred Isaman nfs_request_remove_commit_list(req, &cinfo); 963d6d6dc7cSFred Isaman } 964e824f99aSTrond Myklebust mutex_unlock(&NFS_I(inode)->commit_mutex); 9658dd37758STrond Myklebust nfs_clear_page_commit(req->wb_page); 9668dd37758STrond Myklebust } 967e468bae9STrond Myklebust } 968e468bae9STrond Myklebust 969d45f60c6SWeston Andros Adamson int nfs_write_need_commit(struct nfs_pgio_header *hdr) 9708e821cadSTrond Myklebust { 971c65e6254SWeston Andros Adamson if (hdr->verf.committed == NFS_DATA_SYNC) 972d45f60c6SWeston Andros Adamson return hdr->lseg == NULL; 973c65e6254SWeston Andros Adamson return hdr->verf.committed != NFS_FILE_SYNC; 9748e821cadSTrond Myklebust } 9758e821cadSTrond Myklebust 976919e3bd9STrond Myklebust static void nfs_async_write_init(struct nfs_pgio_header *hdr) 977919e3bd9STrond Myklebust { 978919e3bd9STrond Myklebust nfs_io_completion_get(hdr->io_completion); 979919e3bd9STrond Myklebust } 980919e3bd9STrond Myklebust 981061ae2edSFred Isaman static void nfs_write_completion(struct nfs_pgio_header *hdr) 9826c75dc0dSFred Isaman { 983ea2cf228SFred Isaman struct nfs_commit_info cinfo; 9846c75dc0dSFred Isaman unsigned long bytes = 0; 9856c75dc0dSFred Isaman 9866c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_REDO, &hdr->flags)) 9876c75dc0dSFred Isaman goto out; 988ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, hdr->inode); 9896c75dc0dSFred Isaman while (!list_empty(&hdr->pages)) { 9906c75dc0dSFred Isaman struct nfs_page *req = nfs_list_entry(hdr->pages.next); 9916c75dc0dSFred Isaman 9926c75dc0dSFred Isaman bytes += req->wb_bytes; 9936c75dc0dSFred Isaman nfs_list_remove_request(req); 9946c75dc0dSFred Isaman if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) && 9956c75dc0dSFred Isaman (hdr->good_bytes < bytes)) { 996d1182b33STrond Myklebust nfs_set_pageerror(req->wb_page); 9976c75dc0dSFred Isaman nfs_context_set_write_error(req->wb_context, hdr->error); 9986c75dc0dSFred Isaman goto remove_req; 9996c75dc0dSFred Isaman } 1000c65e6254SWeston Andros Adamson if (nfs_write_need_commit(hdr)) { 1001f79d06f5SAnna Schumaker memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf)); 1002b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, hdr->lseg, &cinfo, 1003a7d42ddbSWeston Andros Adamson hdr->pgio_mirror_idx); 10046c75dc0dSFred Isaman goto next; 10056c75dc0dSFred Isaman } 10066c75dc0dSFred Isaman remove_req: 10076c75dc0dSFred Isaman nfs_inode_remove_request(req); 10086c75dc0dSFred Isaman next: 100920633f04SWeston Andros Adamson nfs_end_page_writeback(req); 10103aff4ebbSTrond Myklebust nfs_release_request(req); 10116c75dc0dSFred Isaman } 10126c75dc0dSFred Isaman out: 1013919e3bd9STrond Myklebust nfs_io_completion_put(hdr->io_completion); 10146c75dc0dSFred Isaman hdr->release(hdr); 10156c75dc0dSFred Isaman } 10166c75dc0dSFred Isaman 1017ce59515cSAnna Schumaker unsigned long 1018ea2cf228SFred Isaman nfs_reqs_to_commit(struct nfs_commit_info *cinfo) 1019fb8a1f11STrond Myklebust { 10205cb953d4STrond Myklebust return atomic_long_read(&cinfo->mds->ncommit); 1021fb8a1f11STrond Myklebust } 1022fb8a1f11STrond Myklebust 1023e824f99aSTrond Myklebust /* NFS_I(cinfo->inode)->commit_mutex held by caller */ 10241763da12SFred Isaman int 1025ea2cf228SFred Isaman nfs_scan_commit_list(struct list_head *src, struct list_head *dst, 1026ea2cf228SFred Isaman struct nfs_commit_info *cinfo, int max) 1027d6d6dc7cSFred Isaman { 1028137da553STrond Myklebust struct nfs_page *req, *tmp; 1029d6d6dc7cSFred Isaman int ret = 0; 1030d6d6dc7cSFred Isaman 1031137da553STrond Myklebust restart: 1032137da553STrond Myklebust list_for_each_entry_safe(req, tmp, src, wb_list) { 10337ad84aa9STrond Myklebust kref_get(&req->wb_kref); 10342ce209c4STrond Myklebust if (!nfs_lock_request(req)) { 10352ce209c4STrond Myklebust int status; 1036137da553STrond Myklebust 1037137da553STrond Myklebust /* Prevent deadlock with nfs_lock_and_join_requests */ 1038137da553STrond Myklebust if (!list_empty(dst)) { 1039137da553STrond Myklebust nfs_release_request(req); 1040137da553STrond Myklebust continue; 1041137da553STrond Myklebust } 1042137da553STrond Myklebust /* Ensure we make progress to prevent livelock */ 10432ce209c4STrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 10442ce209c4STrond Myklebust status = nfs_wait_on_request(req); 10452ce209c4STrond Myklebust nfs_release_request(req); 10462ce209c4STrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 10472ce209c4STrond Myklebust if (status < 0) 10482ce209c4STrond Myklebust break; 1049137da553STrond Myklebust goto restart; 10502ce209c4STrond Myklebust } 1051ea2cf228SFred Isaman nfs_request_remove_commit_list(req, cinfo); 10525d2a9d9dSTrond Myklebust clear_bit(PG_COMMIT_TO_DS, &req->wb_flags); 10538dd37758STrond Myklebust nfs_list_add_request(req, dst); 1054d6d6dc7cSFred Isaman ret++; 10551763da12SFred Isaman if ((ret == max) && !cinfo->dreq) 1056d6d6dc7cSFred Isaman break; 1057e824f99aSTrond Myklebust cond_resched(); 1058d6d6dc7cSFred Isaman } 1059d6d6dc7cSFred Isaman return ret; 1060d6d6dc7cSFred Isaman } 10615d2a9d9dSTrond Myklebust EXPORT_SYMBOL_GPL(nfs_scan_commit_list); 1062d6d6dc7cSFred Isaman 10631da177e4SLinus Torvalds /* 10641da177e4SLinus Torvalds * nfs_scan_commit - Scan an inode for commit requests 10651da177e4SLinus Torvalds * @inode: NFS inode to scan 1066ea2cf228SFred Isaman * @dst: mds destination list 1067ea2cf228SFred Isaman * @cinfo: mds and ds lists of reqs ready to commit 10681da177e4SLinus Torvalds * 10691da177e4SLinus Torvalds * Moves requests from the inode's 'commit' request list. 10701da177e4SLinus Torvalds * The requests are *not* checked to ensure that they form a contiguous set. 10711da177e4SLinus Torvalds */ 10721763da12SFred Isaman int 1073ea2cf228SFred Isaman nfs_scan_commit(struct inode *inode, struct list_head *dst, 1074ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 10751da177e4SLinus Torvalds { 1076d6d6dc7cSFred Isaman int ret = 0; 1077fb8a1f11STrond Myklebust 10785cb953d4STrond Myklebust if (!atomic_long_read(&cinfo->mds->ncommit)) 10795cb953d4STrond Myklebust return 0; 1080e824f99aSTrond Myklebust mutex_lock(&NFS_I(cinfo->inode)->commit_mutex); 10815cb953d4STrond Myklebust if (atomic_long_read(&cinfo->mds->ncommit) > 0) { 10828dd37758STrond Myklebust const int max = INT_MAX; 1083d6d6dc7cSFred Isaman 1084ea2cf228SFred Isaman ret = nfs_scan_commit_list(&cinfo->mds->list, dst, 1085ea2cf228SFred Isaman cinfo, max); 1086ea2cf228SFred Isaman ret += pnfs_scan_commit_lists(inode, cinfo, max - ret); 1087d6d6dc7cSFred Isaman } 1088e824f99aSTrond Myklebust mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex); 1089ff778d02STrond Myklebust return ret; 10901da177e4SLinus Torvalds } 1091d6d6dc7cSFred Isaman 10921da177e4SLinus Torvalds /* 1093e7d39069STrond Myklebust * Search for an existing write request, and attempt to update 1094e7d39069STrond Myklebust * it to reflect a new dirty region on a given page. 10951da177e4SLinus Torvalds * 1096e7d39069STrond Myklebust * If the attempt fails, then the existing request is flushed out 1097e7d39069STrond Myklebust * to disk. 10981da177e4SLinus Torvalds */ 1099e7d39069STrond Myklebust static struct nfs_page *nfs_try_to_update_request(struct inode *inode, 1100e7d39069STrond Myklebust struct page *page, 1101e7d39069STrond Myklebust unsigned int offset, 1102e7d39069STrond Myklebust unsigned int bytes) 11031da177e4SLinus Torvalds { 1104e7d39069STrond Myklebust struct nfs_page *req; 1105e7d39069STrond Myklebust unsigned int rqend; 1106e7d39069STrond Myklebust unsigned int end; 11071da177e4SLinus Torvalds int error; 1108277459d2STrond Myklebust 1109e7d39069STrond Myklebust end = offset + bytes; 1110e7d39069STrond Myklebust 1111f6032f21STrond Myklebust req = nfs_lock_and_join_requests(page); 1112f6032f21STrond Myklebust if (IS_ERR_OR_NULL(req)) 1113f6032f21STrond Myklebust return req; 11142bfc6e56SWeston Andros Adamson 1115e7d39069STrond Myklebust rqend = req->wb_offset + req->wb_bytes; 1116e7d39069STrond Myklebust /* 1117e7d39069STrond Myklebust * Tell the caller to flush out the request if 1118e7d39069STrond Myklebust * the offsets are non-contiguous. 1119e7d39069STrond Myklebust * Note: nfs_flush_incompatible() will already 1120e7d39069STrond Myklebust * have flushed out requests having wrong owners. 1121e7d39069STrond Myklebust */ 1122f6032f21STrond Myklebust if (offset > rqend || end < req->wb_offset) 1123e7d39069STrond Myklebust goto out_flushme; 1124e7d39069STrond Myklebust 11251da177e4SLinus Torvalds /* Okay, the request matches. Update the region */ 11261da177e4SLinus Torvalds if (offset < req->wb_offset) { 11271da177e4SLinus Torvalds req->wb_offset = offset; 11281da177e4SLinus Torvalds req->wb_pgbase = offset; 11291da177e4SLinus Torvalds } 11301da177e4SLinus Torvalds if (end > rqend) 11311da177e4SLinus Torvalds req->wb_bytes = end - req->wb_offset; 1132e7d39069STrond Myklebust else 1133e7d39069STrond Myklebust req->wb_bytes = rqend - req->wb_offset; 1134e7d39069STrond Myklebust return req; 1135e7d39069STrond Myklebust out_flushme: 1136f6032f21STrond Myklebust /* 1137f6032f21STrond Myklebust * Note: we mark the request dirty here because 1138f6032f21STrond Myklebust * nfs_lock_and_join_requests() cannot preserve 1139f6032f21STrond Myklebust * commit flags, so we have to replay the write. 1140f6032f21STrond Myklebust */ 1141f6032f21STrond Myklebust nfs_mark_request_dirty(req); 1142f6032f21STrond Myklebust nfs_unlock_and_release_request(req); 1143e7d39069STrond Myklebust error = nfs_wb_page(inode, page); 1144f6032f21STrond Myklebust return (error < 0) ? ERR_PTR(error) : NULL; 1145e7d39069STrond Myklebust } 11461da177e4SLinus Torvalds 1147e7d39069STrond Myklebust /* 1148e7d39069STrond Myklebust * Try to update an existing write request, or create one if there is none. 1149e7d39069STrond Myklebust * 1150e7d39069STrond Myklebust * Note: Should always be called with the Page Lock held to prevent races 1151e7d39069STrond Myklebust * if we have to add a new request. Also assumes that the caller has 1152e7d39069STrond Myklebust * already called nfs_flush_incompatible() if necessary. 1153e7d39069STrond Myklebust */ 1154e7d39069STrond Myklebust static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx, 1155e7d39069STrond Myklebust struct page *page, unsigned int offset, unsigned int bytes) 1156e7d39069STrond Myklebust { 1157d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 1158e7d39069STrond Myklebust struct nfs_page *req; 1159e7d39069STrond Myklebust 1160e7d39069STrond Myklebust req = nfs_try_to_update_request(inode, page, offset, bytes); 1161e7d39069STrond Myklebust if (req != NULL) 1162e7d39069STrond Myklebust goto out; 11632bfc6e56SWeston Andros Adamson req = nfs_create_request(ctx, page, NULL, offset, bytes); 1164e7d39069STrond Myklebust if (IS_ERR(req)) 1165e7d39069STrond Myklebust goto out; 1166d6d6dc7cSFred Isaman nfs_inode_add_request(inode, req); 1167efc91ed0STrond Myklebust out: 116861e930a9STrond Myklebust return req; 11691da177e4SLinus Torvalds } 11701da177e4SLinus Torvalds 1171e7d39069STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page, 1172e7d39069STrond Myklebust unsigned int offset, unsigned int count) 1173e7d39069STrond Myklebust { 1174e7d39069STrond Myklebust struct nfs_page *req; 1175e7d39069STrond Myklebust 1176e7d39069STrond Myklebust req = nfs_setup_write_request(ctx, page, offset, count); 1177e7d39069STrond Myklebust if (IS_ERR(req)) 1178e7d39069STrond Myklebust return PTR_ERR(req); 1179e7d39069STrond Myklebust /* Update file length */ 1180e7d39069STrond Myklebust nfs_grow_file(page, offset, count); 1181d72ddcbaSWeston Andros Adamson nfs_mark_uptodate(req); 1182a6305ddbSTrond Myklebust nfs_mark_request_dirty(req); 11831d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 1184e7d39069STrond Myklebust return 0; 1185e7d39069STrond Myklebust } 1186e7d39069STrond Myklebust 11871da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page) 11881da177e4SLinus Torvalds { 1189cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 11902a369153STrond Myklebust struct nfs_lock_context *l_ctx; 1191bd61e0a9SJeff Layton struct file_lock_context *flctx = file_inode(file)->i_flctx; 11921da177e4SLinus Torvalds struct nfs_page *req; 11931a54533eSTrond Myklebust int do_flush, status; 11941da177e4SLinus Torvalds /* 11951da177e4SLinus Torvalds * Look for a request corresponding to this page. If there 11961da177e4SLinus Torvalds * is one, and it belongs to another file, we flush it out 11971da177e4SLinus Torvalds * before we try to copy anything into the page. Do this 11981da177e4SLinus Torvalds * due to the lack of an ACCESS-type call in NFSv2. 11991da177e4SLinus Torvalds * Also do the same if we find a request from an existing 12001da177e4SLinus Torvalds * dropped page. 12011da177e4SLinus Torvalds */ 12021a54533eSTrond Myklebust do { 120384d3a9a9SWeston Andros Adamson req = nfs_page_find_head_request(page); 12041a54533eSTrond Myklebust if (req == NULL) 12051a54533eSTrond Myklebust return 0; 12062a369153STrond Myklebust l_ctx = req->wb_lock_context; 1207138a2935STrond Myklebust do_flush = req->wb_page != page || 1208138a2935STrond Myklebust !nfs_match_open_context(req->wb_context, ctx); 1209bd61e0a9SJeff Layton if (l_ctx && flctx && 1210bd61e0a9SJeff Layton !(list_empty_careful(&flctx->flc_posix) && 1211bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_flock))) { 1212d51fdb87SNeilBrown do_flush |= l_ctx->lockowner != current->files; 12135263e31eSJeff Layton } 12141da177e4SLinus Torvalds nfs_release_request(req); 12151a54533eSTrond Myklebust if (!do_flush) 12161a54533eSTrond Myklebust return 0; 1217d56b4ddfSMel Gorman status = nfs_wb_page(page_file_mapping(page)->host, page); 12181a54533eSTrond Myklebust } while (status == 0); 12191a54533eSTrond Myklebust return status; 12201da177e4SLinus Torvalds } 12211da177e4SLinus Torvalds 12221da177e4SLinus Torvalds /* 1223dc24826bSAndy Adamson * Avoid buffered writes when a open context credential's key would 1224dc24826bSAndy Adamson * expire soon. 1225dc24826bSAndy Adamson * 1226dc24826bSAndy Adamson * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL. 1227dc24826bSAndy Adamson * 1228dc24826bSAndy Adamson * Return 0 and set a credential flag which triggers the inode to flush 1229dc24826bSAndy Adamson * and performs NFS_FILE_SYNC writes if the key will expired within 1230dc24826bSAndy Adamson * RPC_KEY_EXPIRE_TIMEO. 1231dc24826bSAndy Adamson */ 1232dc24826bSAndy Adamson int 1233dc24826bSAndy Adamson nfs_key_timeout_notify(struct file *filp, struct inode *inode) 1234dc24826bSAndy Adamson { 1235dc24826bSAndy Adamson struct nfs_open_context *ctx = nfs_file_open_context(filp); 1236dc24826bSAndy Adamson 1237ddf529eeSNeilBrown if (nfs_ctx_key_to_expire(ctx, inode) && 1238ddf529eeSNeilBrown !ctx->ll_cred) 1239ddf529eeSNeilBrown /* Already expired! */ 1240ddf529eeSNeilBrown return -EACCES; 1241ddf529eeSNeilBrown return 0; 1242dc24826bSAndy Adamson } 1243dc24826bSAndy Adamson 1244dc24826bSAndy Adamson /* 1245dc24826bSAndy Adamson * Test if the open context credential key is marked to expire soon. 1246dc24826bSAndy Adamson */ 1247ce52914eSScott Mayhew bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode) 1248dc24826bSAndy Adamson { 1249ce52914eSScott Mayhew struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth; 1250ddf529eeSNeilBrown struct rpc_cred *cred = ctx->ll_cred; 1251ddf529eeSNeilBrown struct auth_cred acred = { 1252*a52458b4SNeilBrown .cred = ctx->cred, 1253ddf529eeSNeilBrown }; 1254ce52914eSScott Mayhew 1255ddf529eeSNeilBrown if (cred && !cred->cr_ops->crmatch(&acred, cred, 0)) { 1256ddf529eeSNeilBrown put_rpccred(cred); 1257ddf529eeSNeilBrown ctx->ll_cred = NULL; 1258ddf529eeSNeilBrown cred = NULL; 1259ddf529eeSNeilBrown } 1260ddf529eeSNeilBrown if (!cred) 1261ddf529eeSNeilBrown cred = auth->au_ops->lookup_cred(auth, &acred, 0); 1262ddf529eeSNeilBrown if (!cred || IS_ERR(cred)) 1263ddf529eeSNeilBrown return true; 1264ddf529eeSNeilBrown ctx->ll_cred = cred; 1265ddf529eeSNeilBrown return !!(cred->cr_ops->crkey_timeout && 1266ddf529eeSNeilBrown cred->cr_ops->crkey_timeout(cred)); 1267dc24826bSAndy Adamson } 1268dc24826bSAndy Adamson 1269dc24826bSAndy Adamson /* 12705d47a356STrond Myklebust * If the page cache is marked as unsafe or invalid, then we can't rely on 12715d47a356STrond Myklebust * the PageUptodate() flag. In this case, we will need to turn off 12725d47a356STrond Myklebust * write optimisations that depend on the page contents being correct. 12735d47a356STrond Myklebust */ 12748d197a56STrond Myklebust static bool nfs_write_pageuptodate(struct page *page, struct inode *inode) 12755d47a356STrond Myklebust { 1276d529ef83SJeff Layton struct nfs_inode *nfsi = NFS_I(inode); 1277d529ef83SJeff Layton 12788d197a56STrond Myklebust if (nfs_have_delegated_attributes(inode)) 12798d197a56STrond Myklebust goto out; 128018dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) 1281d529ef83SJeff Layton return false; 12824db72b40SJeff Layton smp_rmb(); 1283d529ef83SJeff Layton if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags)) 12848d197a56STrond Myklebust return false; 12858d197a56STrond Myklebust out: 128618dd78c4SScott Mayhew if (nfsi->cache_validity & NFS_INO_INVALID_DATA) 128718dd78c4SScott Mayhew return false; 12888d197a56STrond Myklebust return PageUptodate(page) != 0; 12895d47a356STrond Myklebust } 12905d47a356STrond Myklebust 12915263e31eSJeff Layton static bool 12925263e31eSJeff Layton is_whole_file_wrlock(struct file_lock *fl) 12935263e31eSJeff Layton { 12945263e31eSJeff Layton return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX && 12955263e31eSJeff Layton fl->fl_type == F_WRLCK; 12965263e31eSJeff Layton } 12975263e31eSJeff Layton 1298c7559663SScott Mayhew /* If we know the page is up to date, and we're not using byte range locks (or 1299c7559663SScott Mayhew * if we have the whole file locked for writing), it may be more efficient to 1300c7559663SScott Mayhew * extend the write to cover the entire page in order to avoid fragmentation 1301c7559663SScott Mayhew * inefficiencies. 1302c7559663SScott Mayhew * 1303263b4509SScott Mayhew * If the file is opened for synchronous writes then we can just skip the rest 1304263b4509SScott Mayhew * of the checks. 1305c7559663SScott Mayhew */ 1306c7559663SScott Mayhew static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode) 1307c7559663SScott Mayhew { 13085263e31eSJeff Layton int ret; 13095263e31eSJeff Layton struct file_lock_context *flctx = inode->i_flctx; 13105263e31eSJeff Layton struct file_lock *fl; 13115263e31eSJeff Layton 1312c7559663SScott Mayhew if (file->f_flags & O_DSYNC) 1313c7559663SScott Mayhew return 0; 1314263b4509SScott Mayhew if (!nfs_write_pageuptodate(page, inode)) 1315263b4509SScott Mayhew return 0; 1316c7559663SScott Mayhew if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) 1317c7559663SScott Mayhew return 1; 1318bd61e0a9SJeff Layton if (!flctx || (list_empty_careful(&flctx->flc_flock) && 1319bd61e0a9SJeff Layton list_empty_careful(&flctx->flc_posix))) 13208fa4592aSTrond Myklebust return 1; 13215263e31eSJeff Layton 13225263e31eSJeff Layton /* Check to see if there are whole file write locks */ 13235263e31eSJeff Layton ret = 0; 13246109c850SJeff Layton spin_lock(&flctx->flc_lock); 1325bd61e0a9SJeff Layton if (!list_empty(&flctx->flc_posix)) { 1326bd61e0a9SJeff Layton fl = list_first_entry(&flctx->flc_posix, struct file_lock, 1327bd61e0a9SJeff Layton fl_list); 1328bd61e0a9SJeff Layton if (is_whole_file_wrlock(fl)) 13295263e31eSJeff Layton ret = 1; 1330bd61e0a9SJeff Layton } else if (!list_empty(&flctx->flc_flock)) { 13315263e31eSJeff Layton fl = list_first_entry(&flctx->flc_flock, struct file_lock, 13325263e31eSJeff Layton fl_list); 13335263e31eSJeff Layton if (fl->fl_type == F_WRLCK) 13345263e31eSJeff Layton ret = 1; 13355263e31eSJeff Layton } 13366109c850SJeff Layton spin_unlock(&flctx->flc_lock); 13375263e31eSJeff Layton return ret; 1338c7559663SScott Mayhew } 1339c7559663SScott Mayhew 13405d47a356STrond Myklebust /* 13411da177e4SLinus Torvalds * Update and possibly write a cached page of an NFS file. 13421da177e4SLinus Torvalds * 13431da177e4SLinus Torvalds * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad 13441da177e4SLinus Torvalds * things with a page scheduled for an RPC call (e.g. invalidate it). 13451da177e4SLinus Torvalds */ 13461da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page, 13471da177e4SLinus Torvalds unsigned int offset, unsigned int count) 13481da177e4SLinus Torvalds { 1349cd3758e3STrond Myklebust struct nfs_open_context *ctx = nfs_file_open_context(file); 1350d56b4ddfSMel Gorman struct inode *inode = page_file_mapping(page)->host; 13511da177e4SLinus Torvalds int status = 0; 13521da177e4SLinus Torvalds 135391d5b470SChuck Lever nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE); 135491d5b470SChuck Lever 13556de1472fSAl Viro dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n", 13566de1472fSAl Viro file, count, (long long)(page_file_offset(page) + offset)); 13571da177e4SLinus Torvalds 1358149a4fddSBenjamin Coddington if (!count) 1359149a4fddSBenjamin Coddington goto out; 1360149a4fddSBenjamin Coddington 1361c7559663SScott Mayhew if (nfs_can_extend_write(file, page, inode)) { 136249a70f27STrond Myklebust count = max(count + offset, nfs_page_length(page)); 13631da177e4SLinus Torvalds offset = 0; 13641da177e4SLinus Torvalds } 13651da177e4SLinus Torvalds 1366e21195a7STrond Myklebust status = nfs_writepage_setup(ctx, page, offset, count); 136703fa9e84STrond Myklebust if (status < 0) 136803fa9e84STrond Myklebust nfs_set_pageerror(page); 136959b7c05fSTrond Myklebust else 137059b7c05fSTrond Myklebust __set_page_dirty_nobuffers(page); 1371149a4fddSBenjamin Coddington out: 137248186c7dSChuck Lever dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n", 13731da177e4SLinus Torvalds status, (long long)i_size_read(inode)); 13741da177e4SLinus Torvalds return status; 13751da177e4SLinus Torvalds } 13761da177e4SLinus Torvalds 13773ff7576dSTrond Myklebust static int flush_task_priority(int how) 13781da177e4SLinus Torvalds { 13791da177e4SLinus Torvalds switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) { 13801da177e4SLinus Torvalds case FLUSH_HIGHPRI: 13811da177e4SLinus Torvalds return RPC_PRIORITY_HIGH; 13821da177e4SLinus Torvalds case FLUSH_LOWPRI: 13831da177e4SLinus Torvalds return RPC_PRIORITY_LOW; 13841da177e4SLinus Torvalds } 13851da177e4SLinus Torvalds return RPC_PRIORITY_NORMAL; 13861da177e4SLinus Torvalds } 13871da177e4SLinus Torvalds 1388d45f60c6SWeston Andros Adamson static void nfs_initiate_write(struct nfs_pgio_header *hdr, 1389d45f60c6SWeston Andros Adamson struct rpc_message *msg, 1390abde71f4STom Haynes const struct nfs_rpc_ops *rpc_ops, 13911ed26f33SAnna Schumaker struct rpc_task_setup *task_setup_data, int how) 13921da177e4SLinus Torvalds { 13933ff7576dSTrond Myklebust int priority = flush_task_priority(how); 13941da177e4SLinus Torvalds 13951ed26f33SAnna Schumaker task_setup_data->priority = priority; 1396fb91fb0eSAnna Schumaker rpc_ops->write_setup(hdr, msg, &task_setup_data->rpc_client); 13978224b273SChuck Lever trace_nfs_initiate_write(hdr->inode, hdr->io_start, hdr->good_bytes, 13988224b273SChuck Lever hdr->args.stable); 1399275acaafSTrond Myklebust } 1400275acaafSTrond Myklebust 14016d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and 14026d884e8fSFred * call this on each, which will prepare them to be retried on next 14036d884e8fSFred * writeback using standard nfs. 14046d884e8fSFred */ 14056d884e8fSFred static void nfs_redirty_request(struct nfs_page *req) 14066d884e8fSFred { 14076d884e8fSFred nfs_mark_request_dirty(req); 1408c7070113STrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 140920633f04SWeston Andros Adamson nfs_end_page_writeback(req); 14103aff4ebbSTrond Myklebust nfs_release_request(req); 14116d884e8fSFred } 14126d884e8fSFred 1413061ae2edSFred Isaman static void nfs_async_write_error(struct list_head *head) 14146c75dc0dSFred Isaman { 14156c75dc0dSFred Isaman struct nfs_page *req; 14166c75dc0dSFred Isaman 14176c75dc0dSFred Isaman while (!list_empty(head)) { 14186c75dc0dSFred Isaman req = nfs_list_entry(head->next); 14196c75dc0dSFred Isaman nfs_list_remove_request(req); 14206c75dc0dSFred Isaman nfs_redirty_request(req); 14216c75dc0dSFred Isaman } 14226c75dc0dSFred Isaman } 14236c75dc0dSFred Isaman 1424dc602dd7STrond Myklebust static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr) 1425dc602dd7STrond Myklebust { 1426dc602dd7STrond Myklebust nfs_async_write_error(&hdr->pages); 14277be7b3caSTrond Myklebust filemap_fdatawrite_range(hdr->inode->i_mapping, hdr->args.offset, 14287be7b3caSTrond Myklebust hdr->args.offset + hdr->args.count - 1); 1429dc602dd7STrond Myklebust } 1430dc602dd7STrond Myklebust 1431061ae2edSFred Isaman static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = { 1432919e3bd9STrond Myklebust .init_hdr = nfs_async_write_init, 1433061ae2edSFred Isaman .error_cleanup = nfs_async_write_error, 1434061ae2edSFred Isaman .completion = nfs_write_completion, 1435dc602dd7STrond Myklebust .reschedule_io = nfs_async_write_reschedule_io, 1436061ae2edSFred Isaman }; 1437061ae2edSFred Isaman 143857208fa7SBryan Schumaker void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio, 1439a20c93e3SChristoph Hellwig struct inode *inode, int ioflags, bool force_mds, 1440061ae2edSFred Isaman const struct nfs_pgio_completion_ops *compl_ops) 14411751c363STrond Myklebust { 1442a20c93e3SChristoph Hellwig struct nfs_server *server = NFS_SERVER(inode); 144341d8d5b7SAnna Schumaker const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops; 1444a20c93e3SChristoph Hellwig 1445a20c93e3SChristoph Hellwig #ifdef CONFIG_NFS_V4_1 1446a20c93e3SChristoph Hellwig if (server->pnfs_curr_ld && !force_mds) 1447a20c93e3SChristoph Hellwig pg_ops = server->pnfs_curr_ld->pg_write_ops; 1448a20c93e3SChristoph Hellwig #endif 14494a0de55cSAnna Schumaker nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops, 14503bde7afdSTrond Myklebust server->wsize, ioflags); 14511751c363STrond Myklebust } 1452ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_pageio_init_write); 14531751c363STrond Myklebust 1454dce81290STrond Myklebust void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio) 1455dce81290STrond Myklebust { 1456a7d42ddbSWeston Andros Adamson struct nfs_pgio_mirror *mirror; 1457a7d42ddbSWeston Andros Adamson 14586f29b9bbSKinglong Mee if (pgio->pg_ops && pgio->pg_ops->pg_cleanup) 14596f29b9bbSKinglong Mee pgio->pg_ops->pg_cleanup(pgio); 14606f29b9bbSKinglong Mee 146141d8d5b7SAnna Schumaker pgio->pg_ops = &nfs_pgio_rw_ops; 1462a7d42ddbSWeston Andros Adamson 1463a7d42ddbSWeston Andros Adamson nfs_pageio_stop_mirroring(pgio); 1464a7d42ddbSWeston Andros Adamson 1465a7d42ddbSWeston Andros Adamson mirror = &pgio->pg_mirrors[0]; 1466a7d42ddbSWeston Andros Adamson mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize; 1467dce81290STrond Myklebust } 14681f945357STrond Myklebust EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds); 1469dce81290STrond Myklebust 14701da177e4SLinus Torvalds 14710b7c0153SFred Isaman void nfs_commit_prepare(struct rpc_task *task, void *calldata) 14720b7c0153SFred Isaman { 14730b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 14740b7c0153SFred Isaman 14750b7c0153SFred Isaman NFS_PROTO(data->inode)->commit_rpc_prepare(task, data); 14760b7c0153SFred Isaman } 14770b7c0153SFred Isaman 14781f2edbe3STrond Myklebust /* 14791f2edbe3STrond Myklebust * Special version of should_remove_suid() that ignores capabilities. 14801f2edbe3STrond Myklebust */ 14811f2edbe3STrond Myklebust static int nfs_should_remove_suid(const struct inode *inode) 14821f2edbe3STrond Myklebust { 14831f2edbe3STrond Myklebust umode_t mode = inode->i_mode; 14841f2edbe3STrond Myklebust int kill = 0; 1485788e7a89STrond Myklebust 14861f2edbe3STrond Myklebust /* suid always must be killed */ 14871f2edbe3STrond Myklebust if (unlikely(mode & S_ISUID)) 14881f2edbe3STrond Myklebust kill = ATTR_KILL_SUID; 14891f2edbe3STrond Myklebust 14901f2edbe3STrond Myklebust /* 14911f2edbe3STrond Myklebust * sgid without any exec bits is just a mandatory locking mark; leave 14921f2edbe3STrond Myklebust * it alone. If some exec bits are set, it's a real sgid; kill it. 14931f2edbe3STrond Myklebust */ 14941f2edbe3STrond Myklebust if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 14951f2edbe3STrond Myklebust kill |= ATTR_KILL_SGID; 14961f2edbe3STrond Myklebust 14971f2edbe3STrond Myklebust if (unlikely(kill && S_ISREG(mode))) 14981f2edbe3STrond Myklebust return kill; 14991f2edbe3STrond Myklebust 15001f2edbe3STrond Myklebust return 0; 15011f2edbe3STrond Myklebust } 1502788e7a89STrond Myklebust 1503a08a8cd3STrond Myklebust static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr, 1504a08a8cd3STrond Myklebust struct nfs_fattr *fattr) 1505a08a8cd3STrond Myklebust { 1506a08a8cd3STrond Myklebust struct nfs_pgio_args *argp = &hdr->args; 1507a08a8cd3STrond Myklebust struct nfs_pgio_res *resp = &hdr->res; 15082b83d3deSTrond Myklebust u64 size = argp->offset + resp->count; 1509a08a8cd3STrond Myklebust 1510a08a8cd3STrond Myklebust if (!(fattr->valid & NFS_ATTR_FATTR_SIZE)) 15112b83d3deSTrond Myklebust fattr->size = size; 15122b83d3deSTrond Myklebust if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) { 15132b83d3deSTrond Myklebust fattr->valid &= ~NFS_ATTR_FATTR_SIZE; 1514a08a8cd3STrond Myklebust return; 15152b83d3deSTrond Myklebust } 15162b83d3deSTrond Myklebust if (size != fattr->size) 1517a08a8cd3STrond Myklebust return; 1518a08a8cd3STrond Myklebust /* Set attribute barrier */ 1519a08a8cd3STrond Myklebust nfs_fattr_set_barrier(fattr); 15202b83d3deSTrond Myklebust /* ...and update size */ 15212b83d3deSTrond Myklebust fattr->valid |= NFS_ATTR_FATTR_SIZE; 1522a08a8cd3STrond Myklebust } 1523a08a8cd3STrond Myklebust 1524a08a8cd3STrond Myklebust void nfs_writeback_update_inode(struct nfs_pgio_header *hdr) 1525a08a8cd3STrond Myklebust { 15262b83d3deSTrond Myklebust struct nfs_fattr *fattr = &hdr->fattr; 1527a08a8cd3STrond Myklebust struct inode *inode = hdr->inode; 1528a08a8cd3STrond Myklebust 1529a08a8cd3STrond Myklebust spin_lock(&inode->i_lock); 1530a08a8cd3STrond Myklebust nfs_writeback_check_extend(hdr, fattr); 1531a08a8cd3STrond Myklebust nfs_post_op_update_inode_force_wcc_locked(inode, fattr); 1532a08a8cd3STrond Myklebust spin_unlock(&inode->i_lock); 1533a08a8cd3STrond Myklebust } 1534a08a8cd3STrond Myklebust EXPORT_SYMBOL_GPL(nfs_writeback_update_inode); 1535a08a8cd3STrond Myklebust 15361da177e4SLinus Torvalds /* 15371da177e4SLinus Torvalds * This function is called when the WRITE call is complete. 15381da177e4SLinus Torvalds */ 1539d45f60c6SWeston Andros Adamson static int nfs_writeback_done(struct rpc_task *task, 1540d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr, 15410eecb214SAnna Schumaker struct inode *inode) 15421da177e4SLinus Torvalds { 1543788e7a89STrond Myklebust int status; 15441da177e4SLinus Torvalds 1545f551e44fSChuck Lever /* 1546f551e44fSChuck Lever * ->write_done will attempt to use post-op attributes to detect 1547f551e44fSChuck Lever * conflicting writes by other clients. A strict interpretation 1548f551e44fSChuck Lever * of close-to-open would allow us to continue caching even if 1549f551e44fSChuck Lever * another writer had changed the file, but some applications 1550f551e44fSChuck Lever * depend on tighter cache coherency when writing. 1551f551e44fSChuck Lever */ 1552d45f60c6SWeston Andros Adamson status = NFS_PROTO(inode)->write_done(task, hdr); 1553788e7a89STrond Myklebust if (status != 0) 15540eecb214SAnna Schumaker return status; 15558224b273SChuck Lever 1556d45f60c6SWeston Andros Adamson nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count); 15578224b273SChuck Lever trace_nfs_writeback_done(inode, task->tk_status, 15588224b273SChuck Lever hdr->args.offset, hdr->res.verf); 155991d5b470SChuck Lever 1560d45f60c6SWeston Andros Adamson if (hdr->res.verf->committed < hdr->args.stable && 1561d45f60c6SWeston Andros Adamson task->tk_status >= 0) { 15621da177e4SLinus Torvalds /* We tried a write call, but the server did not 15631da177e4SLinus Torvalds * commit data to stable storage even though we 15641da177e4SLinus Torvalds * requested it. 15651da177e4SLinus Torvalds * Note: There is a known bug in Tru64 < 5.0 in which 15661da177e4SLinus Torvalds * the server reports NFS_DATA_SYNC, but performs 15671da177e4SLinus Torvalds * NFS_FILE_SYNC. We therefore implement this checking 15681da177e4SLinus Torvalds * as a dprintk() in order to avoid filling syslog. 15691da177e4SLinus Torvalds */ 15701da177e4SLinus Torvalds static unsigned long complain; 15711da177e4SLinus Torvalds 1572a69aef14SFred Isaman /* Note this will print the MDS for a DS write */ 15731da177e4SLinus Torvalds if (time_before(complain, jiffies)) { 15741da177e4SLinus Torvalds dprintk("NFS: faulty NFS server %s:" 15751da177e4SLinus Torvalds " (committed = %d) != (stable = %d)\n", 1576cd841605SFred Isaman NFS_SERVER(inode)->nfs_client->cl_hostname, 1577d45f60c6SWeston Andros Adamson hdr->res.verf->committed, hdr->args.stable); 15781da177e4SLinus Torvalds complain = jiffies + 300 * HZ; 15791da177e4SLinus Torvalds } 15801da177e4SLinus Torvalds } 15811f2edbe3STrond Myklebust 15821f2edbe3STrond Myklebust /* Deal with the suid/sgid bit corner case */ 158316e14375STrond Myklebust if (nfs_should_remove_suid(inode)) { 158416e14375STrond Myklebust spin_lock(&inode->i_lock); 158516e14375STrond Myklebust NFS_I(inode)->cache_validity |= NFS_INO_INVALID_OTHER; 158616e14375STrond Myklebust spin_unlock(&inode->i_lock); 158716e14375STrond Myklebust } 15880eecb214SAnna Schumaker return 0; 15890eecb214SAnna Schumaker } 15900eecb214SAnna Schumaker 15910eecb214SAnna Schumaker /* 15920eecb214SAnna Schumaker * This function is called when the WRITE call is complete. 15930eecb214SAnna Schumaker */ 1594d45f60c6SWeston Andros Adamson static void nfs_writeback_result(struct rpc_task *task, 1595d45f60c6SWeston Andros Adamson struct nfs_pgio_header *hdr) 15960eecb214SAnna Schumaker { 1597d45f60c6SWeston Andros Adamson struct nfs_pgio_args *argp = &hdr->args; 1598d45f60c6SWeston Andros Adamson struct nfs_pgio_res *resp = &hdr->res; 15991f2edbe3STrond Myklebust 16001f2edbe3STrond Myklebust if (resp->count < argp->count) { 16011da177e4SLinus Torvalds static unsigned long complain; 16021da177e4SLinus Torvalds 16036c75dc0dSFred Isaman /* This a short write! */ 1604d45f60c6SWeston Andros Adamson nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE); 160591d5b470SChuck Lever 16061da177e4SLinus Torvalds /* Has the server at least made some progress? */ 16076c75dc0dSFred Isaman if (resp->count == 0) { 16086c75dc0dSFred Isaman if (time_before(complain, jiffies)) { 16096c75dc0dSFred Isaman printk(KERN_WARNING 16106c75dc0dSFred Isaman "NFS: Server wrote zero bytes, expected %u.\n", 16116c75dc0dSFred Isaman argp->count); 16126c75dc0dSFred Isaman complain = jiffies + 300 * HZ; 16136c75dc0dSFred Isaman } 1614d45f60c6SWeston Andros Adamson nfs_set_pgio_error(hdr, -EIO, argp->offset); 16156c75dc0dSFred Isaman task->tk_status = -EIO; 16166c75dc0dSFred Isaman return; 16176c75dc0dSFred Isaman } 1618f8417b48SKinglong Mee 1619f8417b48SKinglong Mee /* For non rpc-based layout drivers, retry-through-MDS */ 1620f8417b48SKinglong Mee if (!task->tk_ops) { 1621f8417b48SKinglong Mee hdr->pnfs_error = -EAGAIN; 1622f8417b48SKinglong Mee return; 1623f8417b48SKinglong Mee } 1624f8417b48SKinglong Mee 16251da177e4SLinus Torvalds /* Was this an NFSv2 write or an NFSv3 stable write? */ 16261da177e4SLinus Torvalds if (resp->verf->committed != NFS_UNSTABLE) { 16271da177e4SLinus Torvalds /* Resend from where the server left off */ 1628d45f60c6SWeston Andros Adamson hdr->mds_offset += resp->count; 16291da177e4SLinus Torvalds argp->offset += resp->count; 16301da177e4SLinus Torvalds argp->pgbase += resp->count; 16311da177e4SLinus Torvalds argp->count -= resp->count; 16321da177e4SLinus Torvalds } else { 16331da177e4SLinus Torvalds /* Resend as a stable write in order to avoid 16341da177e4SLinus Torvalds * headaches in the case of a server crash. 16351da177e4SLinus Torvalds */ 16361da177e4SLinus Torvalds argp->stable = NFS_FILE_SYNC; 16371da177e4SLinus Torvalds } 1638d00c5d43STrond Myklebust rpc_restart_call_prepare(task); 16391da177e4SLinus Torvalds } 16401da177e4SLinus Torvalds } 16411da177e4SLinus Torvalds 1642af7cf057STrond Myklebust static int wait_on_commit(struct nfs_mds_commit_info *cinfo) 164371d0a611STrond Myklebust { 1644723c921eSPeter Zijlstra return wait_var_event_killable(&cinfo->rpcs_out, 1645723c921eSPeter Zijlstra !atomic_read(&cinfo->rpcs_out)); 1646af7cf057STrond Myklebust } 1647af7cf057STrond Myklebust 1648af7cf057STrond Myklebust static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo) 1649af7cf057STrond Myklebust { 1650af7cf057STrond Myklebust atomic_inc(&cinfo->rpcs_out); 1651af7cf057STrond Myklebust } 1652af7cf057STrond Myklebust 1653af7cf057STrond Myklebust static void nfs_commit_end(struct nfs_mds_commit_info *cinfo) 1654af7cf057STrond Myklebust { 1655af7cf057STrond Myklebust if (atomic_dec_and_test(&cinfo->rpcs_out)) 1656723c921eSPeter Zijlstra wake_up_var(&cinfo->rpcs_out); 165771d0a611STrond Myklebust } 165871d0a611STrond Myklebust 16590b7c0153SFred Isaman void nfs_commitdata_release(struct nfs_commit_data *data) 16601da177e4SLinus Torvalds { 16610b7c0153SFred Isaman put_nfs_open_context(data->context); 16620b7c0153SFred Isaman nfs_commit_free(data); 16631da177e4SLinus Torvalds } 1664e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_commitdata_release); 16651da177e4SLinus Torvalds 16660b7c0153SFred Isaman int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data, 1667c36aae9aSPeng Tao const struct nfs_rpc_ops *nfs_ops, 16689ace33cdSFred Isaman const struct rpc_call_ops *call_ops, 16699f0ec176SAndy Adamson int how, int flags) 16701da177e4SLinus Torvalds { 167107737691STrond Myklebust struct rpc_task *task; 16729ace33cdSFred Isaman int priority = flush_task_priority(how); 1673bdc7f021STrond Myklebust struct rpc_message msg = { 1674bdc7f021STrond Myklebust .rpc_argp = &data->args, 1675bdc7f021STrond Myklebust .rpc_resp = &data->res, 16769ace33cdSFred Isaman .rpc_cred = data->cred, 1677bdc7f021STrond Myklebust }; 167884115e1cSTrond Myklebust struct rpc_task_setup task_setup_data = { 167907737691STrond Myklebust .task = &data->task, 16809ace33cdSFred Isaman .rpc_client = clnt, 1681bdc7f021STrond Myklebust .rpc_message = &msg, 16829ace33cdSFred Isaman .callback_ops = call_ops, 168384115e1cSTrond Myklebust .callback_data = data, 1684101070caSTrond Myklebust .workqueue = nfsiod_workqueue, 16859f0ec176SAndy Adamson .flags = RPC_TASK_ASYNC | flags, 16863ff7576dSTrond Myklebust .priority = priority, 168784115e1cSTrond Myklebust }; 1688788e7a89STrond Myklebust /* Set up the initial task struct. */ 1689e9ae1ee2SAnna Schumaker nfs_ops->commit_setup(data, &msg, &task_setup_data.rpc_client); 16908224b273SChuck Lever trace_nfs_initiate_commit(data); 16911da177e4SLinus Torvalds 1692b4839ebeSKinglong Mee dprintk("NFS: initiated commit call\n"); 1693bdc7f021STrond Myklebust 169407737691STrond Myklebust task = rpc_run_task(&task_setup_data); 1695dbae4c73STrond Myklebust if (IS_ERR(task)) 1696dbae4c73STrond Myklebust return PTR_ERR(task); 1697d2224e7aSJeff Layton if (how & FLUSH_SYNC) 1698d2224e7aSJeff Layton rpc_wait_for_completion_task(task); 169907737691STrond Myklebust rpc_put_task(task); 1700dbae4c73STrond Myklebust return 0; 17011da177e4SLinus Torvalds } 1702e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_initiate_commit); 17031da177e4SLinus Torvalds 1704378520b8SPeng Tao static loff_t nfs_get_lwb(struct list_head *head) 1705378520b8SPeng Tao { 1706378520b8SPeng Tao loff_t lwb = 0; 1707378520b8SPeng Tao struct nfs_page *req; 1708378520b8SPeng Tao 1709378520b8SPeng Tao list_for_each_entry(req, head, wb_list) 1710378520b8SPeng Tao if (lwb < (req_offset(req) + req->wb_bytes)) 1711378520b8SPeng Tao lwb = req_offset(req) + req->wb_bytes; 1712378520b8SPeng Tao 1713378520b8SPeng Tao return lwb; 1714378520b8SPeng Tao } 1715378520b8SPeng Tao 17161da177e4SLinus Torvalds /* 17179ace33cdSFred Isaman * Set up the argument/result storage required for the RPC call. 17189ace33cdSFred Isaman */ 17190b7c0153SFred Isaman void nfs_init_commit(struct nfs_commit_data *data, 1720988b6dceSFred Isaman struct list_head *head, 1721f453a54aSFred Isaman struct pnfs_layout_segment *lseg, 1722f453a54aSFred Isaman struct nfs_commit_info *cinfo) 17239ace33cdSFred Isaman { 17249ace33cdSFred Isaman struct nfs_page *first = nfs_list_entry(head->next); 17252b0143b5SDavid Howells struct inode *inode = d_inode(first->wb_context->dentry); 17269ace33cdSFred Isaman 17279ace33cdSFred Isaman /* Set up the RPC argument and reply structs 17289ace33cdSFred Isaman * NB: take care not to mess about with data->commit et al. */ 17299ace33cdSFred Isaman 17309ace33cdSFred Isaman list_splice_init(head, &data->pages); 17319ace33cdSFred Isaman 17329ace33cdSFred Isaman data->inode = inode; 17339ace33cdSFred Isaman data->cred = first->wb_context->cred; 1734988b6dceSFred Isaman data->lseg = lseg; /* reference transferred */ 1735378520b8SPeng Tao /* only set lwb for pnfs commit */ 1736378520b8SPeng Tao if (lseg) 1737378520b8SPeng Tao data->lwb = nfs_get_lwb(&data->pages); 17389ace33cdSFred Isaman data->mds_ops = &nfs_commit_ops; 1739f453a54aSFred Isaman data->completion_ops = cinfo->completion_ops; 1740b359f9d0SFred Isaman data->dreq = cinfo->dreq; 17419ace33cdSFred Isaman 17429ace33cdSFred Isaman data->args.fh = NFS_FH(data->inode); 17439ace33cdSFred Isaman /* Note: we always request a commit of the entire inode */ 17449ace33cdSFred Isaman data->args.offset = 0; 17459ace33cdSFred Isaman data->args.count = 0; 17460b7c0153SFred Isaman data->context = get_nfs_open_context(first->wb_context); 17479ace33cdSFred Isaman data->res.fattr = &data->fattr; 17489ace33cdSFred Isaman data->res.verf = &data->verf; 17499ace33cdSFred Isaman nfs_fattr_init(&data->fattr); 17509ace33cdSFred Isaman } 1751e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_init_commit); 17529ace33cdSFred Isaman 1753e0c2b380SFred Isaman void nfs_retry_commit(struct list_head *page_list, 1754ea2cf228SFred Isaman struct pnfs_layout_segment *lseg, 1755b57ff130SWeston Andros Adamson struct nfs_commit_info *cinfo, 1756b57ff130SWeston Andros Adamson u32 ds_commit_idx) 175764bfeb49SFred Isaman { 175864bfeb49SFred Isaman struct nfs_page *req; 175964bfeb49SFred Isaman 176064bfeb49SFred Isaman while (!list_empty(page_list)) { 176164bfeb49SFred Isaman req = nfs_list_entry(page_list->next); 176264bfeb49SFred Isaman nfs_list_remove_request(req); 1763b57ff130SWeston Andros Adamson nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx); 1764487b9b8aSTom Haynes if (!cinfo->dreq) 1765487b9b8aSTom Haynes nfs_clear_page_commit(req->wb_page); 17661d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 176764bfeb49SFred Isaman } 176864bfeb49SFred Isaman } 1769e0c2b380SFred Isaman EXPORT_SYMBOL_GPL(nfs_retry_commit); 177064bfeb49SFred Isaman 1771b20135d0STrond Myklebust static void 1772b20135d0STrond Myklebust nfs_commit_resched_write(struct nfs_commit_info *cinfo, 1773b20135d0STrond Myklebust struct nfs_page *req) 1774b20135d0STrond Myklebust { 1775b20135d0STrond Myklebust __set_page_dirty_nobuffers(req->wb_page); 1776b20135d0STrond Myklebust } 1777b20135d0STrond Myklebust 17789ace33cdSFred Isaman /* 17791da177e4SLinus Torvalds * Commit dirty pages 17801da177e4SLinus Torvalds */ 17811da177e4SLinus Torvalds static int 1782ea2cf228SFred Isaman nfs_commit_list(struct inode *inode, struct list_head *head, int how, 1783ea2cf228SFred Isaman struct nfs_commit_info *cinfo) 17841da177e4SLinus Torvalds { 17850b7c0153SFred Isaman struct nfs_commit_data *data; 17861da177e4SLinus Torvalds 1787ade8febdSWeston Andros Adamson /* another commit raced with us */ 1788ade8febdSWeston Andros Adamson if (list_empty(head)) 1789ade8febdSWeston Andros Adamson return 0; 1790ade8febdSWeston Andros Adamson 1791518662e0SNeilBrown data = nfs_commitdata_alloc(true); 17921da177e4SLinus Torvalds 17931da177e4SLinus Torvalds /* Set up the argument struct */ 1794f453a54aSFred Isaman nfs_init_commit(data, head, NULL, cinfo); 1795f453a54aSFred Isaman atomic_inc(&cinfo->mds->rpcs_out); 1796c36aae9aSPeng Tao return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode), 1797c36aae9aSPeng Tao data->mds_ops, how, 0); 17981da177e4SLinus Torvalds } 17991da177e4SLinus Torvalds 18001da177e4SLinus Torvalds /* 18011da177e4SLinus Torvalds * COMMIT call returned 18021da177e4SLinus Torvalds */ 1803788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata) 18041da177e4SLinus Torvalds { 18050b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18061da177e4SLinus Torvalds 1807a3f565b1SChuck Lever dprintk("NFS: %5u nfs_commit_done (status %d)\n", 18081da177e4SLinus Torvalds task->tk_pid, task->tk_status); 18091da177e4SLinus Torvalds 1810788e7a89STrond Myklebust /* Call the NFS version-specific code */ 1811c0d0e96bSTrond Myklebust NFS_PROTO(data->inode)->commit_done(task, data); 18128224b273SChuck Lever trace_nfs_commit_done(data); 1813c9d8f89dSTrond Myklebust } 1814c9d8f89dSTrond Myklebust 1815f453a54aSFred Isaman static void nfs_commit_release_pages(struct nfs_commit_data *data) 1816c9d8f89dSTrond Myklebust { 1817c9d8f89dSTrond Myklebust struct nfs_page *req; 1818c9d8f89dSTrond Myklebust int status = data->task.tk_status; 1819f453a54aSFred Isaman struct nfs_commit_info cinfo; 1820353db796SNeilBrown struct nfs_server *nfss; 1821788e7a89STrond Myklebust 18221da177e4SLinus Torvalds while (!list_empty(&data->pages)) { 18231da177e4SLinus Torvalds req = nfs_list_entry(data->pages.next); 18241da177e4SLinus Torvalds nfs_list_remove_request(req); 182567911c8fSAnna Schumaker if (req->wb_page) 1826d6d6dc7cSFred Isaman nfs_clear_page_commit(req->wb_page); 18271da177e4SLinus Torvalds 18281e8968c5SNiels de Vos dprintk("NFS: commit (%s/%llu %d@%lld)", 18293d4ff43dSAl Viro req->wb_context->dentry->d_sb->s_id, 18302b0143b5SDavid Howells (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)), 18311da177e4SLinus Torvalds req->wb_bytes, 18321da177e4SLinus Torvalds (long long)req_offset(req)); 1833c9d8f89dSTrond Myklebust if (status < 0) { 1834c9d8f89dSTrond Myklebust nfs_context_set_write_error(req->wb_context, status); 183538a33101SKinglong Mee if (req->wb_page) 18361da177e4SLinus Torvalds nfs_inode_remove_request(req); 1837ddeaa637SJoe Perches dprintk_cont(", error = %d\n", status); 18381da177e4SLinus Torvalds goto next; 18391da177e4SLinus Torvalds } 18401da177e4SLinus Torvalds 18411da177e4SLinus Torvalds /* Okay, COMMIT succeeded, apparently. Check the verifier 18421da177e4SLinus Torvalds * returned by the server against all stored verfs. */ 18438fc3c386STrond Myklebust if (!nfs_write_verifier_cmp(&req->wb_verf, &data->verf.verifier)) { 18441da177e4SLinus Torvalds /* We have a match */ 184538a33101SKinglong Mee if (req->wb_page) 18461da177e4SLinus Torvalds nfs_inode_remove_request(req); 1847ddeaa637SJoe Perches dprintk_cont(" OK\n"); 18481da177e4SLinus Torvalds goto next; 18491da177e4SLinus Torvalds } 18501da177e4SLinus Torvalds /* We have a mismatch. Write the page again */ 1851ddeaa637SJoe Perches dprintk_cont(" mismatch\n"); 18526d884e8fSFred nfs_mark_request_dirty(req); 185305990d1bSTrond Myklebust set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags); 18541da177e4SLinus Torvalds next: 18551d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 18567f1bda44STrond Myklebust /* Latency breaker */ 18577f1bda44STrond Myklebust cond_resched(); 18581da177e4SLinus Torvalds } 1859353db796SNeilBrown nfss = NFS_SERVER(data->inode); 1860353db796SNeilBrown if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) 18610db10944SJan Kara clear_bdi_congested(inode_to_bdi(data->inode), BLK_RW_ASYNC); 1862353db796SNeilBrown 1863f453a54aSFred Isaman nfs_init_cinfo(&cinfo, data->inode, data->dreq); 1864af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 18655917ce84SFred Isaman } 18665917ce84SFred Isaman 18675917ce84SFred Isaman static void nfs_commit_release(void *calldata) 18685917ce84SFred Isaman { 18690b7c0153SFred Isaman struct nfs_commit_data *data = calldata; 18705917ce84SFred Isaman 1871f453a54aSFred Isaman data->completion_ops->completion(data); 1872c9d8f89dSTrond Myklebust nfs_commitdata_release(calldata); 18731da177e4SLinus Torvalds } 1874788e7a89STrond Myklebust 1875788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = { 18760b7c0153SFred Isaman .rpc_call_prepare = nfs_commit_prepare, 1877788e7a89STrond Myklebust .rpc_call_done = nfs_commit_done, 1878788e7a89STrond Myklebust .rpc_release = nfs_commit_release, 1879788e7a89STrond Myklebust }; 18801da177e4SLinus Torvalds 1881f453a54aSFred Isaman static const struct nfs_commit_completion_ops nfs_commit_completion_ops = { 1882f453a54aSFred Isaman .completion = nfs_commit_release_pages, 1883b20135d0STrond Myklebust .resched_write = nfs_commit_resched_write, 1884f453a54aSFred Isaman }; 1885f453a54aSFred Isaman 18861763da12SFred Isaman int nfs_generic_commit_list(struct inode *inode, struct list_head *head, 1887ea2cf228SFred Isaman int how, struct nfs_commit_info *cinfo) 188884c53ab5SFred Isaman { 188984c53ab5SFred Isaman int status; 189084c53ab5SFred Isaman 1891ea2cf228SFred Isaman status = pnfs_commit_list(inode, head, how, cinfo); 189284c53ab5SFred Isaman if (status == PNFS_NOT_ATTEMPTED) 1893ea2cf228SFred Isaman status = nfs_commit_list(inode, head, how, cinfo); 189484c53ab5SFred Isaman return status; 189584c53ab5SFred Isaman } 189684c53ab5SFred Isaman 1897c4f24df9STrond Myklebust static int __nfs_commit_inode(struct inode *inode, int how, 1898c4f24df9STrond Myklebust struct writeback_control *wbc) 18991da177e4SLinus Torvalds { 19001da177e4SLinus Torvalds LIST_HEAD(head); 1901ea2cf228SFred Isaman struct nfs_commit_info cinfo; 190271d0a611STrond Myklebust int may_wait = how & FLUSH_SYNC; 1903c4f24df9STrond Myklebust int ret, nscan; 19041da177e4SLinus Torvalds 1905ea2cf228SFred Isaman nfs_init_cinfo_from_inode(&cinfo, inode); 1906af7cf057STrond Myklebust nfs_commit_begin(cinfo.mds); 1907c4f24df9STrond Myklebust for (;;) { 1908c4f24df9STrond Myklebust ret = nscan = nfs_scan_commit(inode, &head, &cinfo); 1909c4f24df9STrond Myklebust if (ret <= 0) 1910c4f24df9STrond Myklebust break; 1911c4f24df9STrond Myklebust ret = nfs_generic_commit_list(inode, &head, how, &cinfo); 1912c4f24df9STrond Myklebust if (ret < 0) 1913c4f24df9STrond Myklebust break; 1914c4f24df9STrond Myklebust ret = 0; 1915c4f24df9STrond Myklebust if (wbc && wbc->sync_mode == WB_SYNC_NONE) { 1916c4f24df9STrond Myklebust if (nscan < wbc->nr_to_write) 1917c4f24df9STrond Myklebust wbc->nr_to_write -= nscan; 1918c4f24df9STrond Myklebust else 1919c4f24df9STrond Myklebust wbc->nr_to_write = 0; 1920c4f24df9STrond Myklebust } 1921c4f24df9STrond Myklebust if (nscan < INT_MAX) 1922c4f24df9STrond Myklebust break; 1923c4f24df9STrond Myklebust cond_resched(); 1924c4f24df9STrond Myklebust } 1925af7cf057STrond Myklebust nfs_commit_end(cinfo.mds); 1926c4f24df9STrond Myklebust if (ret || !may_wait) 1927c4f24df9STrond Myklebust return ret; 1928c4f24df9STrond Myklebust return wait_on_commit(cinfo.mds); 1929c4f24df9STrond Myklebust } 1930c4f24df9STrond Myklebust 1931c4f24df9STrond Myklebust int nfs_commit_inode(struct inode *inode, int how) 1932c4f24df9STrond Myklebust { 1933c4f24df9STrond Myklebust return __nfs_commit_inode(inode, how, NULL); 19341da177e4SLinus Torvalds } 1935b20135d0STrond Myklebust EXPORT_SYMBOL_GPL(nfs_commit_inode); 19368fc795f7STrond Myklebust 1937ae09c31fSAnna Schumaker int nfs_write_inode(struct inode *inode, struct writeback_control *wbc) 19388fc795f7STrond Myklebust { 1939420e3646STrond Myklebust struct nfs_inode *nfsi = NFS_I(inode); 1940420e3646STrond Myklebust int flags = FLUSH_SYNC; 1941420e3646STrond Myklebust int ret = 0; 19428fc795f7STrond Myklebust 1943c4f24df9STrond Myklebust if (wbc->sync_mode == WB_SYNC_NONE) { 19443236c3e1SJeff Layton /* no commits means nothing needs to be done */ 19455cb953d4STrond Myklebust if (!atomic_long_read(&nfsi->commit_info.ncommit)) 1946c4f24df9STrond Myklebust goto check_requests_outstanding; 19473236c3e1SJeff Layton 1948a00dd6c0SJeff Layton /* Don't commit yet if this is a non-blocking flush and there 1949a00dd6c0SJeff Layton * are a lot of outstanding writes for this mapping. 1950420e3646STrond Myklebust */ 19511a4edf0fSTrond Myklebust if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)) 1952420e3646STrond Myklebust goto out_mark_dirty; 1953420e3646STrond Myklebust 1954a00dd6c0SJeff Layton /* don't wait for the COMMIT response */ 1955420e3646STrond Myklebust flags = 0; 1956a00dd6c0SJeff Layton } 1957a00dd6c0SJeff Layton 1958c4f24df9STrond Myklebust ret = __nfs_commit_inode(inode, flags, wbc); 1959c4f24df9STrond Myklebust if (!ret) { 1960c4f24df9STrond Myklebust if (flags & FLUSH_SYNC) 19618fc795f7STrond Myklebust return 0; 1962c4f24df9STrond Myklebust } else if (atomic_long_read(&nfsi->commit_info.ncommit)) 1963c4f24df9STrond Myklebust goto out_mark_dirty; 1964c4f24df9STrond Myklebust 1965c4f24df9STrond Myklebust check_requests_outstanding: 1966c4f24df9STrond Myklebust if (!atomic_read(&nfsi->commit_info.rpcs_out)) 1967c4f24df9STrond Myklebust return ret; 1968420e3646STrond Myklebust out_mark_dirty: 19698fc795f7STrond Myklebust __mark_inode_dirty(inode, I_DIRTY_DATASYNC); 19708fc795f7STrond Myklebust return ret; 19718fc795f7STrond Myklebust } 197289d77c8fSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_write_inode); 1973863a3c6cSAndy Adamson 1974acdc53b2STrond Myklebust /* 1975837bb1d7STrond Myklebust * Wrapper for filemap_write_and_wait_range() 1976837bb1d7STrond Myklebust * 1977837bb1d7STrond Myklebust * Needed for pNFS in order to ensure data becomes visible to the 1978837bb1d7STrond Myklebust * client. 1979837bb1d7STrond Myklebust */ 1980837bb1d7STrond Myklebust int nfs_filemap_write_and_wait_range(struct address_space *mapping, 1981837bb1d7STrond Myklebust loff_t lstart, loff_t lend) 1982837bb1d7STrond Myklebust { 1983837bb1d7STrond Myklebust int ret; 1984837bb1d7STrond Myklebust 1985837bb1d7STrond Myklebust ret = filemap_write_and_wait_range(mapping, lstart, lend); 1986837bb1d7STrond Myklebust if (ret == 0) 1987837bb1d7STrond Myklebust ret = pnfs_sync_inode(mapping->host, true); 1988837bb1d7STrond Myklebust return ret; 1989837bb1d7STrond Myklebust } 1990837bb1d7STrond Myklebust EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range); 1991837bb1d7STrond Myklebust 1992837bb1d7STrond Myklebust /* 1993acdc53b2STrond Myklebust * flush the inode to disk. 1994acdc53b2STrond Myklebust */ 1995acdc53b2STrond Myklebust int nfs_wb_all(struct inode *inode) 199634901f70STrond Myklebust { 1997f4ce1299STrond Myklebust int ret; 199834901f70STrond Myklebust 1999f4ce1299STrond Myklebust trace_nfs_writeback_inode_enter(inode); 2000f4ce1299STrond Myklebust 20015bb89b47STrond Myklebust ret = filemap_write_and_wait(inode->i_mapping); 20026b196875SChuck Lever if (ret) 20036b196875SChuck Lever goto out; 20045bb89b47STrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20056b196875SChuck Lever if (ret < 0) 20066b196875SChuck Lever goto out; 20075bb89b47STrond Myklebust pnfs_sync_inode(inode, true); 20086b196875SChuck Lever ret = 0; 2009f4ce1299STrond Myklebust 20106b196875SChuck Lever out: 2011f4ce1299STrond Myklebust trace_nfs_writeback_inode_exit(inode, ret); 2012f4ce1299STrond Myklebust return ret; 20131c75950bSTrond Myklebust } 2014ddda8e0aSBryan Schumaker EXPORT_SYMBOL_GPL(nfs_wb_all); 20151c75950bSTrond Myklebust 20161b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page) 20171b3b4a1aSTrond Myklebust { 20181b3b4a1aSTrond Myklebust struct nfs_page *req; 20191b3b4a1aSTrond Myklebust int ret = 0; 20201b3b4a1aSTrond Myklebust 2021ba8b06e6STrond Myklebust wait_on_page_writeback(page); 20223e217045SWeston Andros Adamson 20233e217045SWeston Andros Adamson /* blocking call to cancel all requests and join to a single (head) 20243e217045SWeston Andros Adamson * request */ 20256d17d653STrond Myklebust req = nfs_lock_and_join_requests(page); 20263e217045SWeston Andros Adamson 20273e217045SWeston Andros Adamson if (IS_ERR(req)) { 20283e217045SWeston Andros Adamson ret = PTR_ERR(req); 20293e217045SWeston Andros Adamson } else if (req) { 20303e217045SWeston Andros Adamson /* all requests from this page have been cancelled by 20313e217045SWeston Andros Adamson * nfs_lock_and_join_requests, so just remove the head 20323e217045SWeston Andros Adamson * request from the inode / page_private pointer and 20333e217045SWeston Andros Adamson * release it */ 20341b3b4a1aSTrond Myklebust nfs_inode_remove_request(req); 20351d1afcbcSTrond Myklebust nfs_unlock_and_release_request(req); 20361b3b4a1aSTrond Myklebust } 20373e217045SWeston Andros Adamson 20381b3b4a1aSTrond Myklebust return ret; 20391b3b4a1aSTrond Myklebust } 20401b3b4a1aSTrond Myklebust 20411c75950bSTrond Myklebust /* 20421c75950bSTrond Myklebust * Write back all requests on one page - we do this before reading it. 20431c75950bSTrond Myklebust */ 2044c373fff7STrond Myklebust int nfs_wb_page(struct inode *inode, struct page *page) 20451c75950bSTrond Myklebust { 204629418aa4SMel Gorman loff_t range_start = page_file_offset(page); 204709cbfeafSKirill A. Shutemov loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1); 20487f2f12d9STrond Myklebust struct writeback_control wbc = { 20497f2f12d9STrond Myklebust .sync_mode = WB_SYNC_ALL, 20507f2f12d9STrond Myklebust .nr_to_write = 0, 20517f2f12d9STrond Myklebust .range_start = range_start, 20527f2f12d9STrond Myklebust .range_end = range_end, 20537f2f12d9STrond Myklebust }; 20547f2f12d9STrond Myklebust int ret; 20557f2f12d9STrond Myklebust 2056f4ce1299STrond Myklebust trace_nfs_writeback_page_enter(inode); 2057f4ce1299STrond Myklebust 20580522f6adSTrond Myklebust for (;;) { 2059ba8b06e6STrond Myklebust wait_on_page_writeback(page); 20607f2f12d9STrond Myklebust if (clear_page_dirty_for_io(page)) { 2061c373fff7STrond Myklebust ret = nfs_writepage_locked(page, &wbc); 20627f2f12d9STrond Myklebust if (ret < 0) 20637f2f12d9STrond Myklebust goto out_error; 20640522f6adSTrond Myklebust continue; 20657f2f12d9STrond Myklebust } 2066f4ce1299STrond Myklebust ret = 0; 20670522f6adSTrond Myklebust if (!PagePrivate(page)) 20680522f6adSTrond Myklebust break; 20690522f6adSTrond Myklebust ret = nfs_commit_inode(inode, FLUSH_SYNC); 20707f2f12d9STrond Myklebust if (ret < 0) 20717f2f12d9STrond Myklebust goto out_error; 20727f2f12d9STrond Myklebust } 20737f2f12d9STrond Myklebust out_error: 2074f4ce1299STrond Myklebust trace_nfs_writeback_page_exit(inode, ret); 20757f2f12d9STrond Myklebust return ret; 20761c75950bSTrond Myklebust } 20771c75950bSTrond Myklebust 2078074cc1deSTrond Myklebust #ifdef CONFIG_MIGRATION 2079074cc1deSTrond Myklebust int nfs_migrate_page(struct address_space *mapping, struct page *newpage, 2080a6bc32b8SMel Gorman struct page *page, enum migrate_mode mode) 2081074cc1deSTrond Myklebust { 20822da95652SJeff Layton /* 20832da95652SJeff Layton * If PagePrivate is set, then the page is currently associated with 20842da95652SJeff Layton * an in-progress read or write request. Don't try to migrate it. 20852da95652SJeff Layton * 20862da95652SJeff Layton * FIXME: we could do this in principle, but we'll need a way to ensure 20872da95652SJeff Layton * that we can safely release the inode reference while holding 20882da95652SJeff Layton * the page lock. 20892da95652SJeff Layton */ 20902da95652SJeff Layton if (PagePrivate(page)) 20912da95652SJeff Layton return -EBUSY; 2092074cc1deSTrond Myklebust 20938c209ce7SDavid Howells if (!nfs_fscache_release_page(page, GFP_KERNEL)) 20948c209ce7SDavid Howells return -EBUSY; 2095074cc1deSTrond Myklebust 2096a6bc32b8SMel Gorman return migrate_page(mapping, newpage, page, mode); 2097074cc1deSTrond Myklebust } 2098074cc1deSTrond Myklebust #endif 2099074cc1deSTrond Myklebust 2100f7b422b1SDavid Howells int __init nfs_init_writepagecache(void) 21011da177e4SLinus Torvalds { 21021da177e4SLinus Torvalds nfs_wdata_cachep = kmem_cache_create("nfs_write_data", 21031e7f3a48SWeston Andros Adamson sizeof(struct nfs_pgio_header), 21041da177e4SLinus Torvalds 0, SLAB_HWCACHE_ALIGN, 210520c2df83SPaul Mundt NULL); 21061da177e4SLinus Torvalds if (nfs_wdata_cachep == NULL) 21071da177e4SLinus Torvalds return -ENOMEM; 21081da177e4SLinus Torvalds 210993d2341cSMatthew Dobson nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE, 21101da177e4SLinus Torvalds nfs_wdata_cachep); 21111da177e4SLinus Torvalds if (nfs_wdata_mempool == NULL) 21123dd4765fSJeff Layton goto out_destroy_write_cache; 21131da177e4SLinus Torvalds 21140b7c0153SFred Isaman nfs_cdata_cachep = kmem_cache_create("nfs_commit_data", 21150b7c0153SFred Isaman sizeof(struct nfs_commit_data), 21160b7c0153SFred Isaman 0, SLAB_HWCACHE_ALIGN, 21170b7c0153SFred Isaman NULL); 21180b7c0153SFred Isaman if (nfs_cdata_cachep == NULL) 21193dd4765fSJeff Layton goto out_destroy_write_mempool; 21200b7c0153SFred Isaman 212193d2341cSMatthew Dobson nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT, 21224c100210SYanchuan Nian nfs_cdata_cachep); 21231da177e4SLinus Torvalds if (nfs_commit_mempool == NULL) 21243dd4765fSJeff Layton goto out_destroy_commit_cache; 21251da177e4SLinus Torvalds 212689a09141SPeter Zijlstra /* 212789a09141SPeter Zijlstra * NFS congestion size, scale with available memory. 212889a09141SPeter Zijlstra * 212989a09141SPeter Zijlstra * 64MB: 8192k 213089a09141SPeter Zijlstra * 128MB: 11585k 213189a09141SPeter Zijlstra * 256MB: 16384k 213289a09141SPeter Zijlstra * 512MB: 23170k 213389a09141SPeter Zijlstra * 1GB: 32768k 213489a09141SPeter Zijlstra * 2GB: 46340k 213589a09141SPeter Zijlstra * 4GB: 65536k 213689a09141SPeter Zijlstra * 8GB: 92681k 213789a09141SPeter Zijlstra * 16GB: 131072k 213889a09141SPeter Zijlstra * 213989a09141SPeter Zijlstra * This allows larger machines to have larger/more transfers. 214089a09141SPeter Zijlstra * Limit the default to 256M 214189a09141SPeter Zijlstra */ 214289a09141SPeter Zijlstra nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10); 214389a09141SPeter Zijlstra if (nfs_congestion_kb > 256*1024) 214489a09141SPeter Zijlstra nfs_congestion_kb = 256*1024; 214589a09141SPeter Zijlstra 21461da177e4SLinus Torvalds return 0; 21473dd4765fSJeff Layton 21483dd4765fSJeff Layton out_destroy_commit_cache: 21493dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21503dd4765fSJeff Layton out_destroy_write_mempool: 21513dd4765fSJeff Layton mempool_destroy(nfs_wdata_mempool); 21523dd4765fSJeff Layton out_destroy_write_cache: 21533dd4765fSJeff Layton kmem_cache_destroy(nfs_wdata_cachep); 21543dd4765fSJeff Layton return -ENOMEM; 21551da177e4SLinus Torvalds } 21561da177e4SLinus Torvalds 2157266bee88SDavid Brownell void nfs_destroy_writepagecache(void) 21581da177e4SLinus Torvalds { 21591da177e4SLinus Torvalds mempool_destroy(nfs_commit_mempool); 21603dd4765fSJeff Layton kmem_cache_destroy(nfs_cdata_cachep); 21611da177e4SLinus Torvalds mempool_destroy(nfs_wdata_mempool); 21621a1d92c1SAlexey Dobriyan kmem_cache_destroy(nfs_wdata_cachep); 21631da177e4SLinus Torvalds } 21641da177e4SLinus Torvalds 21654a0de55cSAnna Schumaker static const struct nfs_rw_ops nfs_rw_write_ops = { 21664a0de55cSAnna Schumaker .rw_alloc_header = nfs_writehdr_alloc, 21674a0de55cSAnna Schumaker .rw_free_header = nfs_writehdr_free, 21680eecb214SAnna Schumaker .rw_done = nfs_writeback_done, 21690eecb214SAnna Schumaker .rw_result = nfs_writeback_result, 21701ed26f33SAnna Schumaker .rw_initiate = nfs_initiate_write, 21714a0de55cSAnna Schumaker }; 2172